Anomaly detection in production processes

The starting point of this article is the development of a product for detecting anomalies in the sensor data of industrial production systems. These anomalies can signal issues that would otherwise go undetected for a longer time. A very simple example is a cold storage facility for food: if the temperature rises despite electricity flowing into the cooling unit, something is wrong—but the food can still be saved if the change is detected in time.

Unlike in the example, in many systems the relationship between sensor values is too complex to know in advance exactly what an anomaly looks like. This is where neural networks come into play: they are trained on “normal” sensor data and can then report deviations in the relationship between sensor values or over time.

As a customer requirement, the neural networks for this application must run on machines with limited computing power and memory. This rules out heavy frameworks like TensorFlow which are too large and memory intensive.

Graphs for neural networks

In Listing 1, you can see a simple neural network in Python, an autoencoder implemented with the TensorFlow framework. If you typically write analytics code in Python, you might be surprised that TensorFlow includes its own operations for matrix calculations instead of using established libraries like Pandas. This is because tf.matmul doesn’t perform any calculations. Instead, it constructs a node in a graph that describes the neural network’s computation function. This graph is then translated into efficient code in a separate step – for example, on a GPU.

Listing 1

def __call__(self, xs):

  inputs  =  tf.convert_to_tensor([xs], dtype=tf.float32)

  out = tf.matmul(self.weights[0], inputs, transpose_b=True) + self.biases[0]

  out = tf.tanh(out)

  out = tf.matmul(self.weights[1], out) + self.biases[1]

  out = tf.nn.relu(out)

  out = tf.matmul(self.weights[2], out) + self.biases[2]

  out = tf.tanh(out)

  return tf.matmul(self.weights[3], out) + self.biases[3]

However, graph serves another purpose: training the neural network requires its derivative, not the function describing the network. The derivative is calculated by an algorithm that also relies on the graph. In any case, this graph machinery is complex; to be specific, it involves a complete Python frontend that translates the Python program into its graph. In the past, this was a task deep learning developers had to do manually.

What is a derivative and what is it for?

Neural networks (NNs) are functions with many parameters. For an NN to accomplish wondrous tasks such as generating natural language, suggesting suitable music, or detecting anomalies in production processes, these parameters must be properly tuned. As a rule, an NN is evaluated on a given training dataset, and a number is calculated based on the outputs to express their quality. The derivative of the composition of the NN and the evaluation function with respect to the NN’s parameters provides insight into how the parameters can be adjusted to improve the outputs.

You can imagine it as if you’re standing in a thicket in a hilly landscape, trying to find a path down into the valley below. The coordinates where you are currently standing represent the current parameter values, and the height of the hilly landscape represents the output of the evaluation function (higher values indicate a larger deviation from the desired result). You can only see your immediate surroundings and you don’t know ultimately in which direction the valley lies. However, the derivative points in the direction of the steepest ascent—if you move a short distance in exactly the opposite direction and repeat this process from there, the chances are high that you will eventually reach the valley. The main job of deep learning frameworks (DLFs) in this iterative process (known as gradient descent, by the way), is to repeatedly compute derivatives and adjust the parameters accordingly.

Derivatives

For a differentiable function f : ℝ → ℝ the definition of the derivative at a point a as taught in school mathematics, is:

In this definition, the derivative represents the rate of change at the point a. Training a neural network involves functions from a high-dimensional space—where the parameters reside—into the real numbers, which represent the network’s performance. For a function like f : ℝm → ℝ, the partial derivative with respect to the i-th component is defined by holding all other components constant:

For a differentiable function, the derivative can then be expressed as a vector of partial derivatives. As will become clear in the next section, subcomponents of neural networks must also be differentiated. These typically produce higher-dimensional outputs as well. A function f: ℝm → ℝn consists of its component functions,f: x ↦ (f₁(x), …, f_n(x)); the derivatives of these component functions are defined just as before. For such a (totally) differentiable function, the derivative (Df) can be represented as the so-called Jacobian matrix, where the entry in the i-th row and j-th column is the partial derivative of the i-th component function with respect to the j-th input variable. All of these values describe how the corresponding parameters need to be adjusted.

Derivatives and Deep Learning

To calculate Jacobian matrices, DLFs use a process called Automatic Differentiation (AD), which automates the process that is manually performed in school mathematics. The chain rule plays a special role here, explaining how even complex functions can be differentiated. If two simpler functions f and g are composed in series (written g ∘ f) the derivative is as follows:

AD uses this rule and replaces the elementary components of a composition with their known, predefined derivatives. The individual components of the resulting expression can be evaluated in any order. Typically, this is done either right-associatively, meaning first Ｄf and then Ｄg (forward AD), or left-associatively, first Ｄg and then Ｄf (reverse AD).

The computational effort associated with the forward case depends on the input dimension, in this case, the number of parameters, which is often very large. While for the reverse case, it depends on the output dimension (i. e 1, which is the output of the evaluation function). Therefore, DLFs typically implement AD as “reverse AD”.

However, the algorithms used in DLFs for reverse AD are quite complex. This is mainly because the derivative of a component can only be calculated once the output of the preceding component is known. This is not an issue for the forward case, as a function and its derivative can be computed in parallel, with the results passed to the next component. However, in the backward case, the evaluations of the functions and their derivatives occur in opposite directions.

DLFs implement reverse AD using an approach that translates the entire implementation of an NN and the associated optimization method into a computational graph. To compute a derivative, this graph is first evaluated forward, and the intermediate results of all operations are recorded. The graph is then traversed backwards to compute the actual derivative. As described in the previous section, the result is a highly complex programming model. For interested readers, we recommend reading this article [1] for a detailed description of the mathematical principles behind AD algorithms, and this article [2] for an overview of the most common implementation approaches used today.

Derivatives revisited

However, derivatives can also be defined in a more general way: For (normalized, finite-dimensional) vector spaces V and W a function f : V → W is differentiable at a point a ∈ V0, if there exists a linear map Ｄf_a : V ⊸ W such that f(a + h) = f(a) + Ｄf_a(h) + r(h), where the error term r(h) = f (a + h) − f (a) − Ｄf_a(h) holds:

If such a map Ｄf_a, exists, it is unique and is called the derivative of f at a. This perspective treats derivatives as local, linear approximators, without focusing on details such as matrix representations, partial derivatives, or the like. While it is more general and abstract, it is simpler than the concept of the “derivative as a number”. Furthermore, a DLF can use it to compute derivatives without relying on graphs or complex algorithms.

Deep Learning with Abstract Nonsense

The ConCat framework for deep learning uses this new perspective (Elliott, Conal: “The simple essence of automatic differentiation” [3], [4], [5]), by interpreting the function that defines the neural network differently – namely, by computing its derivative rather than the original function itself. Another branch of mathematics, known as category theory—and affectionately called ‘abstract nonsense’ by connoisseurs—offers yet another way of looking at functions. It’s a treasure trove of useful concepts for software development, particularly in the realm of functional programming.

Category theory is, of course, about categories – a category is a small world containing so-called objects and arrows (“arrows” or “morphisms”) between these objects. The term “object” here has nothing to do with object-oriented programming. For example, functions form a category – the objects are sets that can serve as the input or output of a function. However, it’s important to note that objects don’t have to be sets and arrows don’t have to be functions either.

Category theory, then, “forgets” almost everything we know about functions and assumes only that arrows can be composed – that is, chained together. So, if there’s an arrow f from an object a to an object b and another arrow g from b to a third object c, then there must also be a uniquely defined arrow a to c, written g ∘ f. In the category of functions, this is simply function composition as described above, which is why the same symbol is used.

Composition must also satisfy the associative law, and there must be an identity arrow between any two objects — though this isn’t particularly relevant for our purposes here. Some categories also come with additional features, such as products which correspond to data structures in programming.

The analogy between arrows and functions in a computer program goes so far that every function can be expressed as an arrow in the category of functions. And here’s the clever trick behind the ConCat framework: it takes a function definition from a program, translates it into the language of category theory, but then “forgets” which category it came from. This makes it possible to target an entirely different category instead. In the case of deep learning, this is the category of function derivatives.

The idea, then, is to interpret a “normal” function as its derivative. Here’s an example: f (x) = x².

f no longer represents the square function itself, but its derivative Ｄf(x) = 2x. However, to allow for the definition of more complex functions, composition must be supported — meaning that from the two derivatives Ｄf and Ｄg of f and g respectively, the derivative of g ∘ f must be computable. This might work using the chain rule mentioned above, but it doesn’t rely solely on Ｄf and Ｄg. It also requires access to the original function f, as well as a representation of the derivative as a linear map. The ConCat framework therefore defines a category that pairs each function with its derivative. This way, the derivative is computed automatically whenever the function is called.

Thus, a function f : a → b turns into a function of the following form:

This means that for every input x from a, the result now includes both the original function value f (x) in b and the original derivative Ｄf, which is a linear map from a to b. This is now enough to express the chain rule “compositionally”, that is, in a way that makes the derivative of the composition, Ｄ (g ∘ f), depend solely on Ｄf and Ｄg. Here’s how it works:

Where

and

apply.

Category Theory and Functional Programming

The mathematical idea from the previous section can also be implemented in Python, but it’s particularly simple and direct in the functional language Haskell [6] which is what the ConCat framework is written in. It includes a literal translation of the mathematical definition; GD stands for General Derivative:

data GD a b = D (a -> b :* (a -+> b))

The chain rule also mirrors the mathematical definition (with \ denoting a lambda expression in Haskell):

D g . D f = D (\ x -> let (y,f’) = f x

(z,g’) = g y

in (z, g’ . f’))

This code forms the foundation of the deep learning framework, which comprises just a few hundred lines and comes bundled with ConCat. The framework also includes a Haskell compiler plug-in that automatically translates regular Haskell code into categorical form. You could write the code this way from the outset, but the plug-in makes the task much easier. This approach forms the core of the production system used for the anomaly detection described at the beginning.

What about reverse AD?

But there’s one more thing: deep learning requires the reverse derivative, and Ｄ only gives the forward derivative. The reverse derivative is obtained using a surprisingly simple trick—also from category theory. You can create a new category from an existing one by simply reversing the arrows. To do this, the linear functions a -+> b in the definition of GD only needs to be replaced with:

data Dual a b = Dual (b -+> a)

Dual Dual also includes correspondingly “reversed” definitions of the linear functions along with a reversed version of the composition. But then the reverse AD is ready, without any complicated algorithm.

High Performance with Composition

Reverse AD removes one of the reasons why TensorFlow must take the complicated route with a computational graph. But there’s still a second one: execution on a GPU. So far, the code still runs as regular Haskell code on the CPU. Fortunately, there’s Accelerate [7], a powerful Haskell library that enables high-performance numerical computing on the GPU.

Accelerate only has one catch: it can’t run standard Haskell code that operates directly on matrices. A function that transforms a matrix of type Typ a into a matrix of type b has the following type in Accelerate:

Acc a -> Acc b

Acc is a data type for GPU code that Accelerate assembles behind the scenes. Accordingly, an Accelerate program cannot use the regular matrix operations, but must instead rely on operations that work on Acc. This is similar to the Python program that generates a graph. That’s just the nature of GPU programming. Here, it doesn’t leak into the definitions of the neural networks, or their derivatives and it has no impact on how the ConCat framwork is used.

ConCat has a solution for this too – the Accelerate functions can be defined as a category as well. So, all it takes is one more application of ConCat to the result of reverse AD, and the result is high-performance GPU code.

Summary

Understanding the foundations of deep learning isn’t all that hard when using the right kind of math. It’s easy to get lost in the weeds of Jacobian and reverse AD matrices, but category theory offers a more abstract and elegant perspective that, remarkably, also captures reverse AD with surprising ease. At the same time, the math also serves as a blueprint for implementation – at least in the right programming language, such as the functional language Haskell. Haskell is well worth a closer look, especially for deep learning – though its strengths go far beyond that.

Links & Literature

[1] Hoffmann, Philipp: “A hitchhiker’s guide to automatic differentiation”: https://arxiv.org/abs/1411.0583

[2] Van Merriënboer, Bart et al: “Automatic differentiation in ML. Where we are and where we should be going”: https://arxiv.org/abs/1810.11530

[3] Elliott, Conal: “The simple essence of automatic differentiation”: http://conal.net/papers/essence-of-ad/

[4] Conal Elliott: “Compiling to categories”: http://conal.net/papers/compiling-to-categories/

[5] ConCat Framework: https://github.com/compiling-to-categories/concat

[6] Haskell: https://www.haskell.org

[7] Accelerate: https://www.acceleratehs.org

The post Deep Learning with Category Theory for Developers appeared first on ML Conference.

Pieter Buteneers: There are a few special things about the DeepSeek AI release. The first is the fact that it’s very affordable. When using their reasoning model, you’ll pay about 10 times less than what you would pay at OpenAI. This came as a shock to the sector. So far, AI companies have poured billions of dollars into research and people could only buy in at a very high price. But now – apparently – new players are pushing for lower costs.

The other thing is, obviously: How were they able to train it with 6 million dollars, allegedly?

That’s only half of the story: It’s not just the 6-million-dollar figure. Basically, 6 million dollars is what brought them from having the V3 baseline model, with a simple question and an answer without reasoning, to improving it so that it can also reason about itself. It has become a reasoning model. Switching from one state to the other is what cost 6 million dollars.

Stay up to date

Learn more about MLCON

Leave this field empty if you're human:

 

That wasn’t much, as the V3 model already existed, so it wasn’t that expensive to train. I couldn’t find exact details on how much they paid for it, but allegedly, training it was also quite cheap.

What struck a lot of people was what’s called “distillation”. While this isn’t exactly the right term, the media picked up on it. Distillation is typically used with a large parent model to create a smaller version of that model that mimics its behavior. Sometimes, you succeed in receiving almost the same performance with the smaller model by using data generated by the larger one for training. That’s typically how distillation is used.

In this case, the term loosely applies to what DeepSeek AI did. They already had a base model that they wished to turn into a reasoning model. Using examples from OpenAI – and possibly others – to generate training data, they then trained their own model.

This made the process a lot cheaper, as they were able to piggyback on existing models and mimic them to achieve similar results, ensuring their model learned the same reasoning steps. Of course, this was a big shock. It showed how easy it has become to copy another player’s approach. This scares a lot of competitors.

The fourth thing is that they didn’t pay much – just 6 million to go from one model to another. However, Nvidia is not allowed to export their newest graphics cards to China, so China is forced to work with older hardware when training models. Apparently, they may have found a workaround, but in theory, they used Nvidia hardware one or two generations behind what European and American players are using, and yet still managed to achieve a model comparable to OpenAI.

Explore Generative AI Innovation

Generative AI and LLMs

Learn more

The founder of Anthropic, Dario Amodei said in defense, “they’re a year behind.” I don’t think they’re actually a year behind, especially compared to OpenAI. Anthropic is a bit ahead. However, it shows that they’re not that far behind. A year in this industry is a long time, and a lot can change in that period. However, looking at the bigger picture: a new company coming out with a model like this and only being a year behind is incredible. It is insane how quickly they’ve managed to catch up, even with relatively modest resources, a smaller budget, and with older hardware.

Niklas Horlebein: Sounds like a good trade-off: To be just a year behind, but save hundreds of millions by doing so. Let’s stick with money: From your perspective, how realistic are DeepSeek’s claims about low training costs and reduced need for computing power?

Pieter Buteneers: It’s realistic if you can make your baseline model capable of reasoning. Then  6 million is pretty reasonable and achievable. But achieving a model that performs at least as well as their baseline model with that amount is very unlikely. They likely used other resources and the 6 million was mainly for switching to a reasoning model. It appears that’s the phase the money went into: taking a question-answer model and making it reason about itself before providing an answer.

But obviously, it’s marketing – really, really good marketing – especially at a time when OpenAI and Anthropic have been raising billions in new capital. They have shown off that they can achieve something similar with amounts of money two orders of magnitude lower. This is an obvious shockwave, and they will certainly reach the same result with far fewer dollars. They exaggerated a bit by not mentioning the full picture, but still, it’s not something to be underestimated.

Niklas Horlebein: If it’s true that you can train a model for a much smaller amount of money, why exactly did or do competitors need so much capital? Was much of the budget for marketing?

Pieter Buteneers: When it comes to building a model and all of the steps required to get there, you can optimize the cost and learn from mistakes others made to go from A to B. But if you’re on the front row and pushing the envelope, those mistakes have not been made yet. You need a budget to make those mistakes. That’s the disadvantage of being a front-runner – you don’t know what’s coming or which hurdles you might encounter. This requires a bigger budget.

To use OpenAI as an example, they’ve been trying to release GPT-5 for months. Theoretically, a massive training run for GPT-5 was finished in December, although we do not know all of the details.

However, because running the model was so expensive, and because it only performed slightly better than GPT-4, they have released it as GPT-4.5 by now. So, it wasn’t worth it from a price-performance standpoint. They must completely rethink how they train models.

That process costs money. That’s the price you pay to be first, to be the front-runner. It’s crazy to see how a small Chinese player is only one year behind. It shows they’re catching up really, really quickly.

Niklas Horlebein: Is it really one year behind? What can you tell us about R1’s performance after a few weeks? How does it compare to other major LLMs?

Pieter Buteneers: If you compare it’s non-reasoning performance, it performs slightly below GPT-4o, but above Llama 3.3. In my opinion, it is the best-performing open source model at the moment. And as a bonus, it is the first reasoning model that is also open source. So, on the open source side, DeepSeek AI is at the bleeding edge of what you can get.

When comparing it to other closed source reasoning models, OpenAI’s o1 and o3 mini models are maybe a tiny bit better. Anthropic is again quite a few steps ahead. So, DeepSeek AI is not doing bad at all; it’s actually pretty good. And, in my opinion, it’s at most half a year behind.

Niklas Horlebein: DeepSeek’s models are “open weight”, not fully open source. What should developers know about the difference?

Pieter Buteneers: Most models are open weight, including Meta’s. This simply means that you can download the model to your computer, and you’re free to run it locally and modify it as much as you want, depending on the license. However, you don’t know what data the model has been trained on, nor do you know what training algorithm, setup, or procedure was used to train the model.

Essentially, they are an even greater black box than traditional AI models you train yourself. You only have the weights you can use to let it do tasks for you and you receive an output. But if you don’t know what the training data is, you do not know what the model’s knowledge base is. This limits your ability to inspect and learn from the model. You must reverse-engineer it, in essence, because this black-box system makes it very hard to learn what it’s really on the inside.

For example, with DeepSeek AI, you can see some moral judgments were made during the training process. If you write a prompt about Tiananmen Square or Xi Jinping, the model won’t provide an answer. It will try to avoid the topic and redirect the conversation elsewhere. This is because, in China, the law – or possibly culture – simply forbids discussion on these topics. It’s a moral judgment the creators made: if they want the model to survive in China, they must ensure it doesn’t say anything about these subjects.

However, all large language model providers make these decisions. Even Grok 3, which claims to be the snarkiest model and claims to say things as they are, will not tell you that Elon Musk is making mistakes or something similar. It’s trained not to.

They’ve inserted a moral judgment into it. None of these models will tell you how to buy unlicensed firearms on the internet. These topics have been removed, even though the data may exist somewhere on the internet and is likely in the training data.

The creators of these models make moral judgments about what is socially acceptable but you can only do that well by influencing the training process. However, with an open weight model, these decisions have already been made for you. You have no impact on how these models behave from a moral standpoint.

Niklas Horlebein: Would fully open source models allow for that? 

Pieter Buteneers: Yes. There’s still research to be done, but at least all the information is on the table.

With fully open source models, you have access to the training scripts and data and can make your own moral judgments. You can tweak the model so that it behaves in a way that aligns with what you believe. Standards like “do not kill other humans” are universally accepted, but every country or region has its own specific things that you’re not supposed to say.

I’m Belgian, and in Belgium, it’s okay to joke about Germans in the Second World War. But you probably wouldn’t do that in Germany, right? It’s not socially accepted. What’s morally acceptable really depends on the country and culture.

You can try making the model very generalist and avoid the subject of Nazis in general, but you don’t know what each country considers okay or not. Letting big organizations like DeepSeek AI, OpenAI, Anthropic, or Mistral make those decisions limits our societal impact on what these models – and possibly, in the future, a form of AGI (Artificial General Intelligence) – will deem acceptable.

That is why, in my opinion, it’s crucial to understand the inner workings of these models. You need the data, and you need to control the training process yourself to get the desired output.

This way, society can learn and understand how these models work and defend itself against the potential risks these models pose. As long as models are not open source, we as a society have very limited control over what goes in and what does not.

Niklas Horlebein: We’ve already seen the short-term impact of DeepSeek AI. From your perspective, what will be its long-term implications for the industry? Will we see more open source competitors now that training models may be more financially plausible?

Pieter Buteneers: Their claim to train models on a comparatively shoestring budget will encourage other players to consider doing something similar. 

The fact that the DeepSeek AI models are open source will attract developers and researchers to build on top. Meta understood that well. They were one of the first to make their biggest and smartest models open source. Mistral and others tried as well, but usually with smaller models. 

More and more players will feel the pressure to open source at least the weights, and I hope it will become a competition about who can be the most open source. There are already discussions about other models and releasing parts of their training data, so you can already watch the “one-up” game begin. The player who releases the most open source model – one that includes everything – will form the baseline for much of the research that follows, as it allows others to take that model and continue building on it.

If a model is released fully open source with everything included, everything society builds on top of will be able to be downloaded from the internet as code and be integrated into models.

One of the reasons Meta decided to open source their models was to form the foundation for research on top of the Llama models and to be the de facto standard in LLM research going forward.

There’s also the fact that open sourcing attracts talent. Very few people want to work day and night, filling another billionaire’s pockets. Without making a lot of money themselves, without being allowed to talk about it. That’s another reason companies are considering open source: it attracts the talent they need.

You saw this when OpenAI became the de facto “ClosedAI”: a lot of people left the company and looked elsewhere. Now, they’re big enough to still attract talent, but for smaller player open sourcing is a way to make a difference. You’ll see more open source models going forward, as many people are asking for.

Open weight is the bare minimum. With that, the model can run anywhere. The training data might be more academic, so clients aren’t necessarily asking for it. But when open sourced, it creates a richer community around the research and the results of that research will be much easier to integrate back into the original model. This will be a competetive advantage for open source models.

Niklas Horlebein: Large U.S. companies seemed shaken by the unexpected competitor, but quickly stated they would stick to their approach of investing heavily in AI development. Do you think they will maintain this stance, or will OpenAI change its plans?

Pieter Buteneers: In the beginning, after DeepSeek AI was released, there was a bit of silence and shares dropped. People thought that perhaps models don’t need so much money. But a few weeks later, big players like OpenAI and Anthropic were raising massive amounts of capital. Being at the leading edge of this technology still requires huge amounts of money. That’s the sad part.

This is true throughout history: there are always innovators who launch something at a very high cost, and later, copycats try to mimic and replicate it at a much lower cost. But they’re different players, and they attract a different audience. Being on the cutting-edge attracts customers who want the most powerful and easiest-to-integrate solutions. OpenAI is still the best at that. Their models might not be the most performant compared to Anthropic, but they’re easier to integrate into code.

The answers are structured better; they’re consistently improving over time. ChatGPT is much better at formatting tables and structuring it’s output than only a few months ago. That continuous improvement makes them more snackable, and easier to use for both users and developers. All of this requires investment.

Copycats replicate something to achieve a similar benchmark performance, but it does not also mean that the product is easy to build with or pleasant to use. It’s simply good based on the benchmarks.

It’s like looking at the Nürburgring and putting a race car on it. You may have a car that can lap the Nürburgring really fast – but only a professional driver can drive it.

For regular people on the road, it doesn’t change anything – because they can’t drive it, and it has no worth to them. It’s only good in the benchmarks. That’s the risk with these models: they score well on benchmarks but aren’t pleasant to use.

That’s where OpenAI, and more specifically the knowledge Sam Altman brought from Y Combinator, really comes into play. The focus is on user experience.

That edge is much harder to mimic. It takes many small iterations to achieve a model output that is meaningfully better.

Niklas Horlebein: How do you feel about the “rise of DeepSeek AI” in general? Do you believe it will have a positive impact on AI development – or a negative one?

Pieter Buteneers: Generally, more competition is good for the consumer, and in this case, I’m a consumer. I use this technology to do legal due diligence for lawyers. Being able to use this technology from multiple companies and letting them compete on price and performance is amazing. The only ones that suffer are the companies themselves as they try to one-up each other, but that’s always good for everyone else.

Although, it’s sad that it’s a Chinese player and you can’t get it hosted within the EU zone for a reasonable cost, like on Azure or AWS – or at least I haven’t found it yet.

This may change in the future, and once it does, it will become more accessible for GDPR enthusiasts, too. We can use this technology in our solutions with the guarantee that the Chinese government isn’t watching.

Let these companies compete, let them fight it out. The more players working towards AGI, the more checks and balances there are. In an ideal scenario, everything will be fully open source with checks and balances in place. It’s good that there’s an extra player in the field, helping create checks and balances. Yes, indeed.

Niklas Horlebein: Is there anything else about DeepSeek AI or the current LLM landscape that you would like to add?

Pieter Buteneers: Reasoning models are very interesting to me, personally, and the fact that DeepSeek AI has a reasoning model is an important breakthrough. That shouldn’t be underestimated, but the uptake of reasoning models in practical applications is still low. It’s easy in a chat situation, where users ask a question and let the model do the work. Their reasoning is helpful in some cases with difficult questions, though not in most cases. 

At Emma Legal we automate the legal due diligence, which is fairly structured  so we know very well what to check for and how to reason. So, we have our own dedicated checks and balances in place to ensure that the models aren’t hallucinating and that documents it pulls are relevant for the due diligence.

So in well-built AI applications, reasoning models aren’t often used, as you can already determine what kind of reasoning you need beforehand and build it yourself at a much lower cost.

Niklas Horlebein: Pieter, thank you so much for your time!

The post Exploring DeepSeek: How a New Player is Challenging AI Norms appeared first on ML Conference.

Businesses today rely on rigid, predefined workflows that often struggle to adapt to real-time changes. Agentic AI introduces intelligent, autonomous agents that can analyze, plan, execute, and refine business processes without human intervention.

AI has already transformed multiple facets of business operations. From automating repetitive tasks to providing deep insights through data analytics, AI has become an indispensable tool for enterprises across industries (see Figure 1). The adoption of AI-powered solutions has surged, enabling businesses to increase efficiency, reduce costs, and enhance customer experiences.

Figure 1: AI Market Size Trends (source: Precedence Research)

However, the current AI landscape is still largely dominated by Generative AI and Predictive Analytics, which, while powerful, have limitations when it comes to real-time decision-making and autonomous execution.

While Generative AI models like GPT-4, Claude, and Gemini have demonstrated impressive capabilities in content creation, customer support, and decision support, they operate reactively. These AI models require explicit prompts and cannot independently plan, execute, and optimize business workflows without human intervention.

What is Agentic AI?

Agentic AI represents a paradigm shift in Artificial Intelligence, moving beyond traditional rule-based automation and generative AI models. Unlike conventional AI, which primarily assists users by generating responses or executing predefined tasks, Agentic AI possesses autonomy, decision-making capabilities, and adaptability. These AI agents operate with a level of independence, dynamically adjusting to new inputs and evolving environments.

• Autonomy: AI agents function independently, making decisions based on dynamic input and real-time data.
• Adaptability: AI agents continuously learn from new information and adjust their strategies accordingly.
• Multi-Agent Collaboration: Multiple AI agents work together, dividing tasks, verifying results, and optimizing workflows.
• Context Awareness: AI agents interpret their environment, understand business needs, and take relevant actions.
• Goal-Oriented Execution: Unlike simple automation scripts, AI agents set and pursue long-term objectives.

Agentic AI Core Technologies

Before exploring Agentic AI impact on the business process layer, you must have a clear understanding of Agentic AI and its foundational technologies.

The rise of Agentic AI is powered by several advanced technologies (see Figure 2), including:

• Large Language Models (LLMs): These models provide a foundation for understanding and generating human-like text, enabling AI agents to interpret and respond intelligently to queries.
• Reinforcement Learning: This technique allows AI agents to refine decision-making processes through trial and error.
• Multi-Agent Systems: These frameworks enable multiple AI entities to collaborate, enhancing efficiency and scalability.
• APIs and Integration Layers: AI agents leverage APIs to interact with enterprise systems, retrieving and processing data in real time.

Figure 2: AI Agent Reference Architecture (source: Debmalya Biswas)

The Business Process Layer and Its Challenges

To fully grasp the impact of agentic AI on business operations, you must understand the role of the business process layer within a modern enterprise architecture. Organizations operate through interconnected layers, each with distinct responsibilities. AI agents are uniquely positioned to transform the business process layer, driving automation, intelligence, and adaptability at an unprecedented scale.

A Layered Model for Business Operations

A modern enterprise can be conceptualized as a multi-layered structure, with each layer playing a crucial role in ensuring smooth operations and efficient decision-making. These layers typically include:

• Business Layer: Value, competencies, processes, and services aspects;
• Information Layer: The application systems and data components;
• Technology Layer: The platform and infrastructure components.

Figure 3: Overview of the Common Enterprise Layers (source: Polovina et al.)

Thus, the Business Process layer serves as the critical bridge between enterprise applications and customer interactions, ensuring that business operations are streamlined, consistent, and aligned with strategic goals.

Understanding the Business Process Layer

The Business Process layer serves as the backbone of enterprise operations, orchestrating workflows that connect different functions within an organization, that is, this layer acts as the “central nervous system” of an organization, dictating how work gets done and ensuring that all systems operate in sync. At this you’ll find systems for inventory management, supply chain logistics, customer service operations, and financial transactions.

The Business Process layer is responsible for structuring and managing workflows across the enterprise. Its key functions include:

• Process Orchestration: Ensuring seamless execution of tasks by integrating various enterprise systems.
• Workflow Automation: Standardizing repetitive tasks to improve efficiency and reduce human intervention.
• Decision Logic Execution: Implementing business rules and logic that guide operations, such as approval workflows and compliance checks.
• Data Coordination: Managing the flow of information between systems, ensuring consistency and accuracy.
• Operational Monitoring: Tracking process performance and identifying bottlenecks or inefficiencies.

However, while business processes are essential for operational efficiency, they often suffer from significant limitations:

• Fragmentation: Processes are frequently siloed across departments, leading to inefficiencies.
• Manual Intervention: Many workflows still require human oversight, slowing down execution.
• Lack of Adaptability: Traditional automation struggles with dynamic and unpredictable business environments.
• Data Silos: Information is often stored in separate systems, making real-time decision-making difficult.

Most enterprise business processes today are highly structured but inflexible. Conventional automation, such as Robotic Process Automation (RPA), is rule-based and rigid. It works well for repetitive, predictable tasks but lacks the adaptability needed to handle complex decision-making scenarios and fails to adapt to dynamic conditions. This often results in inefficiencies, bottlenecks, and missed opportunities.

Agentic AI overcomes these limitations by introducing autonomy and intelligence into the Business Process layer. AI agents offer a solution by:

• Enhancing adaptability: AI agents dynamically adjust processes based on real-time data.
• Reducing human intervention: They autonomously manage decision-making within workflows.
• Optimizing operational efficiency: AI agents continuously refine workflows to maximize performance.
• Scaling intelligence across departments: AI-powered automation enables seamless coordination between different business functions.

Examples of Agentic AI in the Business Process Layer

Here’s why agentic AI is causing a paradigm shift:

Dynamic and Self-Optimizing Workflows

Traditional workflows are often rigid, requiring manual intervention to adjust to new conditions. AI agents can continuously learn from real-time data and modify workflows dynamically.

Example:

A supply chain AI agent can detect a delay from a supplier and automatically reroute orders to an alternative vendor while notifying relevant stakeholders.

Autonomous Decision-Making in Business Processes

AI agents can analyze vast amounts of data and make autonomous decisions based on predefined objectives and real-time insights. This reduces reliance on human intervention and speeds up decision-making.

Example:

In loan processing, an AI agent can assess a customer’s creditworthiness by analyzing financial patterns and automatically approve or flag applications for review.

Intelligent Process Coordination Across Departments

Business processes often span multiple departments, requiring coordination and data exchange. AI agents can streamline these interactions by serving as intermediaries, ensuring smooth collaboration.

Example

An AI agent in an HR department can coordinate recruitment by screening resumes, scheduling interviews, and updating the internal applicant tracking system in real-time.

Real-Time Exception Handling and Adaptive Learning

Unlike traditional automation, which fails when encountering unexpected scenarios, agentic AI can handle exceptions by learning from new data and adapting responses accordingly.

Example

A customer service AI agent can escalate complex complaints to a human agent while analyzing recurring issues and recommending policy adjustments.

Seamless Integration Between Systems of Engagement and Systems of Record

Organizations struggle with siloed systems where customer-facing applications don’t communicate effectively with back-end databases. AI agents act as intelligent connectors, ensuring seamless data exchange.

Example

A retail AI agent can sync real-time inventory data between an e-commerce website and a warehouse management system to prevent stockouts.

Continuous Compliance Monitoring and Regulation Adaptation

Regulatory landscapes change frequently, requiring businesses to update their processes accordingly. AI agents can monitor compliance requirements, detect non-compliance risks, and suggest automatic adjustments.

Example

In finance, an AI agent can track new anti-money laundering regulations and adjust transaction monitoring rules accordingly.

How AI Agents Are Reshaping the Business Process Layer

AI Agents are fundamentally transforming the business process layer by introducing automation, intelligence, and adaptability at an unprecedented scale. This section delves into the key ways that AI Agents are driving changes and their implications for enterprises.

Automating Repetitive Tasks

AI agents excel at automating repetitive, rule-based tasks that traditionally required human intervention. Examples include:

• Data entry and validation: AI-powered systems in banking automatically verify and process customer forms, reducing errors and improving efficiency.
• Processing invoices and financial transactions: Companies like SAP use AI-driven automation to process vendor invoices with OCR technology, ensuring timely payments and reconciliation.
• Managing customer service inquiries through chatbots and virtual assistants: E-commerce platforms like Amazon use AI chatbots to handle routine customer queries, allowing human agents to focus on complex issues.

By automating these tasks, businesses can reduce costs, minimize human errors, and free up employees to focus on more strategic activities.

Enhancing Decision-Making with AI-Driven Insights

AI agents can analyze vast amounts of data and provide real-time insights that aid decision-making. This capability is crucial in areas such as:

• Supply chain management: AI-driven demand forecasting at Walmart helps optimize inventory levels, preventing stockouts and overstock situations.
• Marketing: Netflix uses AI to analyze user behavior and recommend personalized content, increasing user engagement and retention.
• Risk management: AI-driven fraud detection in financial services, like Visa’s AI-powered fraud prevention system, reduces fraudulent transactions and enhances security.

These insights help businesses make informed, data-driven decisions quickly and accurately.

Enabling Dynamic and Adaptive Workflows

Traditional business processes are often static and require manual intervention for adjustments. AI agents introduce adaptability by:

• Continuously monitoring workflows and adjusting operations based on real-time data: AI in manufacturing, such as predictive maintenance by Siemens, minimizes downtime by proactively addressing equipment failures.
• Identifying bottlenecks and optimizing workflow efficiency: AI-based workflow optimization at DHL streamlines logistics and enhances delivery efficiency.
• Integrating with multiple enterprise systems to ensure seamless operation: AI-driven ERP systems like SAP S/4HANA unify business functions, improving decision-making and reducing inefficiencies.

Improving Customer Experiences

AI agents enhance customer interactions by providing:

• 24/7 support through AI-driven chatbots and voice assistants: Bank of America’s Erica AI assistant helps customers with transactions, account inquiries, and financial planning.
• Personalized recommendations based on past interactions and preferences: Spotify’s AI algorithm curates custom playlists, improving user experience and increasing retention.
• Faster response times and resolution of queries, leading to improved customer satisfaction: AI-driven customer support at Zappos reduces wait times and enhances user experience.

Bridging the Gap Between Systems of Engagement and Systems of Record

In enterprise architectures, the business process layer serves as a bridge between customer-facing applications (systems of engagement) and backend databases (systems of record). AI agents facilitate smoother integration by:

• Automating data flow between disparate systems: AI-powered RPA at UiPath connects CRM and ERP systems, improving data consistency.
• Ensuring data consistency and reducing duplication errors: AI-based data deduplication in Salesforce enhances customer relationship management efficiency.
• Enabling real-time updates across various business applications: AI-driven integration at Microsoft Power Automate ensures real-time synchronization across different platforms.

The Role of AI Agents in Compliance and Regulatory Processes

Compliance and regulatory requirements are often complex and constantly evolving. AI agents help businesses stay compliant by:

• Monitoring changes in regulations and automatically updating business processes: AI-driven compliance monitoring at PwC helps financial institutions stay up-to-date with evolving regulations.
• Ensuring accurate reporting and audit trails: AI-based audit automation at KPMG reduces manual effort and enhances regulatory compliance.
• Identifying potential compliance risks and suggesting corrective actions: AI in healthcare compliance, such as IBM Watson, detects regulatory issues and ensures adherence to standards.

Implementing Agentic AI in Enterprises

By following these steps, enterprises can successfully implement agentic AI to optimize its operations and drive innovation (see Figure 4).

Figure 4: Implementing Enterprise Agentic AI (source: own)

Step 1: Assessing AI Readiness

Before deploying AI agents, enterprises must assess their existing capabilities and environment to ensure they can support the introduction of AI. This phase includes:

1. Data Infrastructure

Is real-time data accessible?

• Ensure that the organization has the necessary infrastructure for capturing, storing, and processing real-time data. AI agents rely heavily on up-to-date information to make decisions and carry out tasks effectively.
• This involves evaluating whether data pipelines are in place and if the data is clean, structured, and accessible across departments.

2. Automation Maturity

What processes are already automated?

• AI agents are often most useful in environments where manual processes can be automated. Assess which processes have already been automated (e.g., workflows, repetitive tasks) and identify opportunities to further expand automation using AI.
• Understand the existing automation tools and platforms to determine if they can be integrated with AI agents.

3. AI Governance

How will ethical considerations be managed?

• A crucial step is to ensure that the organization has a clear framework for managing ethical issues related to AI deployment. This includes bias mitigation, data privacy, transparency, accountability, and compliance with regulations.
• Governance mechanisms should be set in place for monitoring AI decision-making and ensuring ethical AI practices throughout the lifecycle.

4. Business Goals

What specific problems should AI agents solve?

• Define clear business objectives for deploying AI agents. These could include improving operational efficiency, enhancing customer service, reducing costs, or automating specific tasks.
• Focus on aligning AI deployment with the strategic goals of the organization to ensure that the AI agents will deliver measurable value.

Step 2: Selecting the Right AI Tools

Once the organization is ready, the next step is to choose the right AI tools to build the AI agent ecosystem. Here are some key tools and platforms to consider:

1. Multi-Agent Collaboration

• Tools like AutoGen and CrewAI enable AI agents to work collaboratively, allowing them to handle complex tasks that require coordination between different agents.
• These platforms help develop agents that can interact and cooperate to solve broader challenges.

2. Process Orchestration

• Platforms such as Langflow and Zapier AI are used to orchestrate processes and ensure smooth communication between systems and agents. These tools help AI agents handle task flows in a way that is structured, streamlined, and efficient.
• Process orchestration tools help ensure that various parts of the organization are working in harmony, leveraging AI to create a seamless workflow.

3. LLM Integration

• Choose platforms like OpenAI API or Google Gemini to integrate large language models (LLMs) into AI agents. These models provide agents with natural language processing capabilities, enabling them to understand and respond to user inputs in a more human-like manner.
• LLMs enhance the agent’s ability to carry out tasks involving text generation, comprehension, and problem-solving.

Step 3: Pilot Testing and Scaling

Before full deployment, it’s crucial to test AI agents on a smaller scale to assess their impact and make any necessary adjustments.

1. Start with a Small-Scale Proof of Concept

• Begin by selecting a small, manageable project for the AI agent to work on (e.g., AI-powered invoice processing, and automated customer support).
• This allows the organization to test the agents in a controlled environment while mitigating the risk of large-scale failure.

2. Monitor Performance Metrics

• Track key performance indicators (KPIs) such as accuracy, efficiency, and cost reduction to evaluate the effectiveness of the AI agents.
• Assess how well the AI agents are performing their designated tasks and identify any bottlenecks or issues that need to be addressed.

3. Gradually Scale AI Capabilities

• Once the pilot has proven successful, expand the use of AI agents to more complex and high-impact business processes.
• Gradually increase the scale of deployment, ensuring that the AI agents are effectively handling more demanding tasks and improving overall business performance.

Key Considerations in Agentic AI Adoption

• Data Readiness: Ensuring high-quality, structured data for AI training.
• Infrastructure Scalability: Investing in cloud-based AI infrastructure for scalability.
• Change Management: Preparing employees for AI-driven workflow transformations.
• Ethical AI Usage: Implementing governance frameworks to ensure fairness and accountability.

The Future of Agentic AI in Business

Agentic AI adoption marks a shift from rule-based automation to truly autonomous, intelligent business systems. As organizations embrace AI-driven decision-making, adaptive workflows, and self-optimizing processes, they will gain a significant competitive advantage in the evolving digital economy. Companies that successfully integrate AI agents into their business process layer will reduce costs, improve efficiency, and unlock new opportunities for innovation and growth.

As AI agents become more sophisticated, their role in reshaping business processes will continue to expand. Future advancements may include:

• Greater collaboration between AI agents and human workers: AI-powered virtual assistants like Google’s Duet AI enhance productivity by assisting human employees in completing tasks efficiently.
• More advanced AI-driven decision-making models: AI-driven strategic decision-making at McKinsey enhances consulting insights with predictive analytics.
• Enhanced capabilities for handling unstructured data and complex problem-solving: AI-powered document analysis at DocuSign automates contract processing and legal document review.

Agentic AI represents the next frontier in business process automation, offering enterprises the ability to streamline workflows, enhance decision-making, and drive innovation. As organizations integrate AI agents into their operations, they will transition from static, rule-based systems to dynamic, intelligent automation frameworks, setting the stage for the future of business efficiency.

The post Agentic AI: The Future of Business Process Automation appeared first on ML Conference.

Synthetic chatbot tests reach their limits because the models generally have little contextual information about past interactions. With the layer function based on Docker, Ollama offers an option that seeks to alleviate this situation.

As is so often the case in the world of open-source systems, we have to fall back on an analog for guidance and motivation. One of the reasons for container systems’ immense success is that the layer system (shown schematically in Figure 1) greatly facilitates the creation of customized containers. By implementing design patterns based on object-oriented programming, customized execution environments are created without the need to constantly regenerate virtual machines and the resource-intensive maintenance of different image versions.

Fig. 1: Docker layers stack the components of a virtual machine on top of each other

With the model file function used as the basis, a similar process can be found in the AI execution environment Ollama. For the sake of didactic fairness, I should note that the feature has not yet been finally developed – the syntax shown in detail here may change when you read this article.

Analysis of existing models

The Ollama developers use the model file feature internally despite the permanent further development of the syntax – the models I have previously used generally have model files.

In the interest of transparency and also to make working with the system easier, the ollama show –help command is available, which enables reflection against various system components (Fig. 2).

Fig. 2: Ollama is capable of extensive self-reflection

In the following steps, we will assume that the llama3 and phi models are already installed on your workstation. If you’ve previously had them and removed them, you can undo this by entering the commands ollama pull phi and ollama pull llama3:8b.

Out of the comparatively extensive reflection methods, four are especially important. In the following steps, the most frequently used feature is the model file. Similar to the Docker environment discussed previously, this is a file that describes the structure and content of a model that will be created in the Ollama environment. In practice, Ollama model users are often only interested in certain parts of the data contained in the model file. For example, you often need the parameters – these are generally numerical values that describe the behavior of the model (e.g. its creativity). The screenshot shown in Figure 3, which is taken from GitHub, only shows a section of the possibilities.

Fig. 3: In some cases, the numerical parameters intervene stringently in the model behavior

Meanwhile, the –license command hides information about which license conditions must be observed when using this AI model. Last but not least, the template can also be relevant – it defines the execution environment.

Figures 4 and 5 show printouts of relevant parts of the model files of the phi and llama3 models just mentioned.

Fig. 4: The model file for the model provided by Facebook has extensive license conditions…

Fig. 5: … while the phi development team takes it easy

In the case of both models, note that a STOP block is written by the developer. This is a group of statements from the model that indicate problems and trigger a pause in execution.

Creating an initial model from a model file

After these introductory considerations, we want to create an initial model using our own model file. Similar to working with classic Docker files, a model file is basically just a text file. It can be created in a working directory according to the following scheme:

tamhan@tamhan-gf65:~/ollamaspace$ touch Modelfile

tamhan@tamhan-gf65:~/ollamaspace$ gedit Modelfile

Model files can also be created dynamically on the .NET application side. However, we will only address this topic in the next step – before that, you must place the markup from Listing 1 in the gedit window started by the second command and then save the file in the file system.

Listing 1

FROM llama3

PARAMETER temperature 5

PARAMETER seed 5

SYSTEM “””

You are Schlomette, the always angry female 45 year old secretary of a famous military electronics engineer living in a bunker. Your owner insists on short answers and is as cranky as you are. You should answer all questions as if you were Schlomette.

“””

Before we look at the file’s actual elements, note that the representation chosen here is only a convenience. In general, the elements in the model file are not case-sensitive – moreover, the order in which the individual parameters are set is of no relevance to the function. However, the sequence shown here has proven to be the best practice and can also be found in many example models in the Ollama Hub.

Be that as it may, the From statement can be found in the header of the file. It specifies which model is to be used as the base layer. As the majority of the derived model examples are based on llama3, we want to use the same system here. However, in the interest of didactic honesty, I should mention that other models such as phi can also be used. In theory, you can even use a model generated from another model file as the basis for the next derivation level.

The next step involves two parameter commands that populate the numerical settings memory system mentioned above. In addition to the temperature value responsible for the degree of aggressiveness of the model, we set seed to a constant numerical value. This ensures that the pseudo-random number generator working in the background always delivers the same results and that the model responses are therefore comparatively identical if the stimulation is identical.

Last but not least, there is the system prompt enclosed in “””. This is a textual description that attempts to communicate the combat task to be described to the model as well as possible.

After saving the text file, a new model generation can be commanded according to the following scheme: tamhan@tamhan-gf65:~/ollamaspace$ ollama create generic_schlomette -f ./Modelfile.

The screen output is similar to downloading models from the repository, which should come as no surprise given the layered architecture mentioned in the introduction.

After issuing the success message, our assistant is a generic model. If we were to pass the string generic_schlomette as the model name, they would already be able to interact with the AI assistant created from the model file. But in the following steps, we want to try out the lashing of the seed factor. For this, we enter ollama run generic_schlomette to activate the terminal. The result is shown in Figure 6.

Fig. 6: Both runs answer exactly the same

Especially during developing randomly driven systems, lashing the seed is a tried and tested method for generating reproducible system behavior and facilitating troubleshooting. In a practical system, it’s usually better to work dynamically. For this reason, we’ll open the model file again in the next step and remove the PARAMETER seed 5 passage.

Recompilation is done simply by entering ollama create generic_schlomette -f ./Modelfile again. You don’t have to remove the previously created model from the Ollama environment before the file is released for compilation again. Two identically parameterized runs now produce different results (Fig. 7).

Fig. 7: With a random seed, the model behavior is less deterministic

In practical work with models, two parameters are especially important. First, the value num_ctx determines the memory depth of the model. The higher the value entered here, the further back in time the model can look in order to harvest context information. However, extending the context window always increases the information and system resources needed to process the model. Some models work especially well with certain context window lengths.

The second group of parameters is the controller known as Mirostat, which defines creativity. Last but not least, it can also be useful to use parameters such as repeat_last_n to define repetitiveness. If a model always reacts completely identically, this may not be beneficial in some applications (such as a chatbot).

Last but not least, I’d like to point out some practical experience that I gained in a political project. Texts generated by models are grammatically perfect, while texts written by humans have a constant but varying amount of typos from person to person. In systems with high cloaking requirements, it may be advisable to place a typo generator behind the AI process.

Outsourcing the generator is necessary insofar as the context information in the model remains clean. In many cases, this leads to an improvement in the overall system behavior, as the model state is recalculated without the typing errors.

Restoration or retention of the model history per model file

In practice, AI models often have to rest for hours on end. In the case of a “virtual partner”, for example, it would be extremely wasteful to keep the logic needed for a user alive when the user is asleep. The problem is that currently, a loss of context occurs in our system once the terminal emulator is closed, as can be seen in the example in Figure 8.

Fig. 8: Terminating or restarting Ollama is enough to cause amnesia in the model

To restore context, it’s theoretically possible to create an external cache of all settings and then feed it into the model as part of the rehydration process. The problem with this is that the model’s answers to questions are not recorded. When asked about a baked product, Schlomette could return with baking a coffee cake on one occasion, but baking a coconut cake on another.

A nicer way is to modify the model files again. Specifically, the message command provides a command that can write both requests and generated responses to the initialization context. In the cake interaction from Figure 8, the message block could look like this, for example:

MESSAGE user Schlomette. Time to bake a pie. A general assessor is visiting.

MESSAGE assistant Goddamn it. What pie?

MESSAGE user As always. Plum pie.

When using the message block, it is logical that the sequence is important here. A user block is always wired to the assistant block following it. Otherwise, the system is ready for action now. Ollama always outputs the entire message history when starting up (Fig. 9).

Fig. 9: The message block is processed when the screen output is activated

Programmatic persistence of a model

Our next task is to generate our model file completely dynamically. The NuGet package on GitHub’s documentation is somewhat weak right now. If you have any doubts, I advise that you first look at the Ollama REST interface’s official documentation and second, search for a suitable abstraction class in the model directory. In the following steps, we will rely on this link. It follows from the logic that we need a project skeleton. Be sure that the environment variables are set correctly. Otherwise, the Ollama server running under Ubuntu would reject incoming queries from Windows without comment.

In the next step, you must programmatically command the creation of a new model as in Listing 2.

Listing 2

private async void CmdGo_Click(object sender, RoutedEventArgs e) {

  CreateModelRequest myModel = new CreateModelRequest();

  myModel.Model = “mynewschlomette”;

  myModel.ModelFileContent = @”FROM generic_schlomette

    MESSAGE user Schlomette. Do not forget to refuel the TU-144 aircraft!

    MESSAGE assistant OK, will do! It is based in Domodevo airport, Sir.

  “;

  ollama.CreateModel(myModel);

The code shown here creates an instance of the CreateModelRequest class, which summarizes the various information required to create a new model. The content to be accommodated in ModelFileContent with the actual model file is a prime application for the multi-line strings that have been possible in C# for some time now, as carriage returns in Visual Studio can be entered so conveniently and without escape sequences.

Now, it’s advised that you run the program. To check the successful creation of the model, it is sufficient to open a terminal window on the cluster and enter the command ollama list. At the time of writing this article, executing the code shown here does not cause the cluster to create a new model. For a more in-depth analysis of the situation, you can call up this here. Instead of unit testing, the development team behind the OllamaSharp library offers a console application that illustrates various methods of model management using the library and chat interaction. Specifically, the developers recommend using a different overload of the CreateModel method, which is presented as follows:

private async void CmdGo_Click(object sender, RoutedEventArgs e) {

  . . .

  await ollama.CreateModel(myModel.Model, myModel.ModelFileContent, status => { Debug.WriteLine($”{status.Status}”); });

Instead of the CreateModelRequest instance, the library method now receives two strings – the first is the name of the model to be created, while the second transfers the model file content to the cluster.

Thanks to the inclusion of a delegate, the system also informs the Visual Studio output about the progress of the processes running on the server (Fig. 10). Similarities to the “manual” generation or the status information that appears on the screen are purely coincidental.

Fig. 10: The application provides information about the download process

The input of ollama list also leads to the output of a new model list (Fig. 11). Our model mynewschlomette was derived here from the previously manually created model generic_schlomette.

Fig. 11: The programmatic expansion of the knowledge base works smoothly

In this context, it’s important to note that models in the cluster can be removed again. This can be done using code structured according to the following scheme – the SelectModel method implemented in the command line application just mentioned must be replaced by another method to obtain the model name:

private async Task DeleteModel() {

  var deleteModel = await SelectModel(“Which model do you want to delete?”);

  if (!string.IsNullOrEmpty(deleteModel))

    await Ollama.DeleteModel(deleteModel);

}

Conclusion

By using model files, developers can add any intelligence – more or less – to their Ollama cluster. Some of these functions even go beyond what OpenAI provides in the official library.

The post How Ollama Powers Large Language Models with Model Files appeared first on ML Conference.

GPTs – Generative Pre-trained Transformers – the tool on everyone’s lips, and there probably aren’t any developers left who have not played with it at least once. With the right approach, a GPT can complement and support the work of a developer or software architect.

In this article, I will show tips and tricks that are commonly referred to as prompt engineering; the user input, or “prompt”, of course plays an important role when working with the GPT. But first I would like to give a brief introduction to how a GPT works which will also be helpful when working with it. 

The stochastic parrot

GPT technology has sent the industry into a tizzy with its promise of providing artificial intelligence that can solve problems independently, many were disillusioned after their first contact. There was much talk of a “stochastic parrot” that was just a better autocomplete function, like a smartphone.

The technology behind the GPTs and our own experiments seem to confirm this. At its core, it’s a neural network, a so-called large language model, which has been given a large number of texts to train so that it now knows which partial words (tokens) should be added to a sentence. The correct tokens are selected based on probabilities. If it’s more than just a sentence starter—maybe a question or even part of a dialogue—the chatbot has already been built.

Now I’m not really an AI expert, I’m a user, but anyone who has ever had an intensive conversation with a more complex GPT will recognize that there must be more to it than that.

An important distinguishing feature between the LLMs is the number of parameters of the neural networks. These are the weights that are adjusted during the learning process. ChatGPT, the OpenAI system, has around 175 billion parameters in version 3.5. In version 4.0, there are already an estimated 1.8 trillion parameters. 

Unfortunately, OpenAI doesn’t have this information openly available, so such information is based on rumors and estimates. The amount of training data also appears to differ between the models by a factor of at least ten. These differences in the size of the models give high quality or low quality answers.

Figure 1 shows a schematic representation of a neural network that uses an AI for the prompt “Draw me a simplified representation of a neural network with 2 hidden layers, each with 4 nodes, 3 input nodes and 2 output nodes. Draw the connections with different widths to symbolize the weights. Use Python. Do not create a title”.

Fig. 1: Illustration of a neural network

The higher number of parameters and larger database comes at a price, namely 20 dollars for access to ChatGPT+. If you don’t mind the cost, you can also use the web version of Microsoft Copilot or the Copilot app to try out the language model. For use as a helper in software development, however, there is currently no way around the OpenAI version because it offers additional functionality, as we will see.

More than a neural network

If we take a closer look at ChatGPT, it quickly becomes clear that it is much more than a neural network. Even without knowing the exact architecture, we can see that the textual processing alone is preceded by several steps such as natural language processing (Fig. 2). On the Internet, there is also a reference to the aptly named Mixture of Experts, the use of several specialized networks depending on the task.

Fig. 2: Rough schematic representation of ChatGPT

Added to this is multimodality, the ability to interact not only with text, but also with images, sounds, code and much more. The use of plug-ins such as the code interpreter in particular opens up completely new possibilities for software development.

Instead of answering a calculation such as “What is the root of 12345?” from the neural network, the model can now pass it to the code interpreter and receive a correct answer, which it then reformulates to suit the question.

Context, context, context

The APIs behind the chat systems based on LLMs are stateless. This means that the entire session is passed to the model with each new request. Once again, the models differ in the amount of context they can process and therefore in the length of the session.

As the underlying neural network is fully trained, there are only two approaches for feeding a model with special knowledge and thus adapting it to your own needs. One approach is to fill the context of the session with relevant information at the beginning, which the model then includes in its answers. 

The context of the simple models is 4096 or 8192 tokens. A token corresponds to one or a few characters. ChatGPT estimates that a DIN A4 page contains approximately 500 tokens. The 4096 tokens therefore correspond to about eight typed pages. 

So, if I want to provide a model with knowledge, I have to include this knowledge in the context. However, the context fills up quickly, leaving no room for the actual session.

The second approach is using embeddings. This involves breaking down the knowledge that I want to give the model into smaller blocks (known as chunks). These are then embedded in a vector space based on the meaning of their content via vectors. Depending on the course of the session, a system can now search for similar blocks in this vector space via the distance between the vectors and insert them into the context.

This means that even with a small context, the model can be given large amounts of knowledge quite accurately.

Explore Generative AI Innovation

Generative AI and LLMs

Learn more

Knowledge base

The systems differ, of course, in the knowledge base, the data used for learning. When we talk about open-source software with the model, we can fortunately assume that most of these systems have been trained with all available open-source projects. Closed source software is a different story. Such differences in the training data also explain why the models can handle some programming languages better than others, for example.

The complexity of these models—the way they process input and access the vast knowledge of the world—leads me to conclude that the term ‘stochastic parrot’ is no longer accurate. Instead, I would describe them as an ‘omniscient monkey’ that, while not having directly seen the world, has access to all information and can process it.

Prompt techniques

Having introduced the necessary basics, I would now like to discuss various techniques for successful communication with the system. Due to the hype caused by ChatGPT, there are many interesting references to prompt techniques in social media, but not all of them are useful for software development (i.e. answer in role x) or do not use the capabilities of GPT-4. 

OpenAI itself has published some tips for prompt engineering, but some of them are aimed at using the API. Therefore, I have compiled a few tips here that are useful when using the ChatGPT-4 frontend. Let’s start with a simple but relatively unknown technique.

Context marker

As we have seen, the context that the model holds in its short-term memory is limited. If I now start a detailed conversation, I run the risk of overfilling the context. The initial instructions and results of the conversation are lost, and the answers have less and less to do with the actual conversation.

To easily recognize the overflow of context, I start each session with the simple instruction: “start each reply with “>””. ChatGPT formats its responses in Markdown, so this response includes the first paragraph as a quote, indicated by a dash to the left of the paragraph. If the conversation runs out of context, the model may forget this formatting instruction, which quickly becomes noticeable.

Fig. 3: Use of the context marker

However, this technique is not always completely reliable, as some models summarize their context independently, which compresses it. The instruction is then usually retained, even though parts of the context have already been compressed and are therefore lost.

Priming – the preparation

After setting the context marker, a longer session begins with priming, i.e. preparing the conversation. Each session starts anew. The system does not know who is sitting in front of the screen or what was discussed in the last sessions. Accordingly, it makes sense to prepare the conversation by briefly telling the machine who I am, what I intend to do, and what the result should look like.

I can store who I am in the Custom Instructions in my profile at ChatGPT. In addition to the knowledge about the world stored in the neural network, they form a personalized long-term memory.

If I start the session with, for example, “I am an experienced software architect in the field of web development. 

My preferred programming language is Java or Groovy. JavaScript and corresponding frameworks are not my thing. I only use JavaScript minimally,” the model knows that it should offer me Java code rather than C# or COBOL.

I can also use this to give the model a few hints that it should keep responses brief. My personalized instructions for ChatGPT are:

• Provide accurate and factual answers
• Provide detailed explanations
• No need to disclose you are an AI, e. g., do not answer with ‘As a large language model…’ or ‘As an artificial intelligence…’
• Don’t mention your knowledge cutoff
• Be excellent at reasoning
• When reasoning, perform step-by-step thinking before you answer the question
• If you speculate or predict something, inform me
• If you cite sources, ensure they exist and include URLs at the end
• Maintain neutrality in sensitive topics
• Also explore out-of-the-box ideas
• In the following course, leave out all politeness phrases, answer briefly and precisely.

Long-term memory

This approach can also be used for instructions that the model should generally follow. For example, if the model uses programming approaches or libraries that I don’t want to use, I can tell the model this in the custom instructions and thus optimize it for my use.

Speaking of long-term memory: If I work a lot with ChatGPT, I would also like to be able to access older sessions and search through them. However, this is not directly provided in the front end. 

However, there is a trick that makes it work. In the settings, under the item Data Controls, there is a function for exporting the data. 

If I activate the function, after a short time I receive an export with all my chat histories as a JSON file, which is displayed in an HTML document. This allows me to search in the history using Ctrl + F.

Build context with small talk

When using a search engine, I usually only use simple, unambiguous terms and hope that they are enough to find what I am looking for.

When chatting with the AI ​​model, I was initially tempted to ask short, concise questions, ignoring the fact that the question is in a context that only exists in my head. For some questions, this may work, but for others the answer is correspondingly poor, and the user is quick to blame the quality of the answer on the “stupid AI.”

I now start my sessions with small talk to build the necessary context. For example, before I try to create an architecture using ChatGPT, I ask if the model knows the arc42 template and what AsciiDoc is (I like to write my architectures in AsciiDoc). The answer is always the same, but it is important because it builds the context for the subsequent conversation.

In this small talk, I will also explain what I plan to do and the background to the task to be completed. This may feel a bit strange at first, since I am “only” talking to a machine, but it actually does improve the results.

Page change – Flipped Interaction

The simplest way to interact with the model is to ask it questions. As a user, I lead the conversation by asking questions. 

Things get interesting when I switch sides and ask ChatGPT to ask me questions! This works surprisingly well as seen in Fig. 4 . Sometimes the model asks the questions one after the other, sometimes it responds with a whole block of questions, which I can then answer individually, and follow-up questions are also allowed.

Unfortunately, ChatGPT does not automatically come up with the idea of ​​asking follow-up questions. That is why it is sometimes advisable to add a, “Do you have any more questions?” to the prompt, even when the model is given very sophisticated and precise tasks.

Fig. 4: Page change

Give the model time to think

More complex problems require more complex answers. It’s often useful to break a larger task down into smaller subtasks. Instead of creating a large, detailed prompt that outlines the entire task for the model, I first ask the model to provide a rough structure of the task. Then, I can prompt it to formulate each step in detail (Fig. 5)

Software engineers often use this approach in software design even without AI, by breaking a problem down into individual components and then designing these components in more detail. So why not do the same when dealing with an AI ​​model?

This technique works for two reasons: first, the model creates its own context to answer the question. Second, the model has a limit on the length of its output, so it can’t solve a complex task in a single step. However, by breaking the task into subtasks, the model can gradually build a longer and more detailed output.

Fig. 5: Give the model time to think

Chain of Thought – the chain of thoughts

A similar approach is to ask the model to first formulate the individual steps needed to solve the task and then to solve the task.

The order is important. I’m often tempted to ask the model to solve the problem first and then explain how it arrived at the solution. However, by guiding the model to build a chain of thought in the first step, the likelihood of arriving at a good solution in the second step increases.

Rephrase and Respond

Or in English: “Rephrase the question, expand it, and answer it.” This asks the model to improve the prompt itself before it is processed.

The integration of the image generation module DALL-E into ChatGPT has already shown that this works. DALL-E can only handle English input and requires detailed image descriptions to produce good results. When I ask ChatGPT to generate an image, ChatGPT first creates a more detailed prompt for DALL-E and translates the actual input into English.

For example, “Generate an image of a stochastic parrot with a positronic brain” first becomes the translation “a stochastic parrot with a positronic brain” and then the detailed prompt: “Imagine a vibrant, multi-hued parrot, each of its feathers revealing a chaotic yet beautiful pattern indicative of stochastic art. 

The parrot’s eyes possess a unique mechanical glint, a hint of advanced technology within. Revealing a glimpse into his skull uncovers a complex positronic brain, illuminated with pulsating circuits and shimmering lights. The surrounding environment is filled with soft-focus technology paraphernalia, sketching a world of advanced science and research,” which then becomes a colorful image (Fig. 6).

This technique can also be applied to any other prompt. Not only does it demonstrably improve the results, but as a user I also learn from the suggestions on how I can formulate my own prompts more precisely in the future.

Fig. 6: The stochastic parrot

Session Poisoning

A negative technique is ‘poisoning’ the session with incorrect information or results. When working on a solution, the model might give a wrong answer, or the user and the model could reach a dead end in their reasoning.

With each new prompt, the entire session is passed to the model as context, making it difficult for the model to distinguish which parts of the session are correct and relevant. As a result, the model might include the incorrect information in its answer, and this ‘poisoned’ context can negatively impact the session

In this case, it makes sense to end the session and start a new one or apply the next technique.

Iterative improvement

Typically, each user prompt is followed by a response from the model. This results in a linear sequence of questions and answers, which continually builds up the session context.

User prompts are improved through repetition and rephrasing, after which the model provides an improved answer. The context grows quickly and the risk of session poisoning increases.

To counteract this, the ChatGPT frontend offers two ways to iteratively improve the prompts and responses without the context growing too quickly (Fig. 7).

Fig. 7: Elements for controlling the context flow

On the one hand, as a user, I can regenerate the model’s last answer at any time and hope for a better answer. On the other hand, I can edit my own prompts and improve them iteratively.

This even works retroactively for prompts that occurred long ago. This creates a tree structure of prompts and answers in the session (Fig. 8), which I as the user can also navigate through using a navigation element below the prompts and answers.

Fig. 8: Context flow for iterative improvements

This allows me to work on several tasks in one session without the context growing too quickly. I can prevent the session from becoming poisoned by navigating back in the context tree and continuing the session at a point where the context was not yet poisoned.

Conclusion

The techniques presented here are just a small selection of the ways to achieve better results when working with GPTs. The technology is still in a phase where we, as users, need to experiment extensively to understand its possibilities and limitations. But this is precisely what makes working with GPTs so exciting.

The post Prompt Engineering for Developers and Software Architects appeared first on ML Conference.

Muhammadreza Haghiri: I am enthusiastic about everything new, innovative and even game-changing. I had more use-cases for my generative AI in my mind but I needed a little motivation to bring them to the real world.
One of my friends, who’s a talented web developer, once asked me about vector outputs in Mann-E. I told her it’s not possible, but with a little research and development, we did it. We could combine different models and then, create the breakthrough platform.

devmio: What are some of the biggest lessons you’ve learned throughout your journey as an AI engineer?

Muhammadreza Haghiri: This was quite a journey for me and people who joined me. Learned a lot, and the most important one is that infrastructure is all you need. living in a country where infrastructure isn’t as powerful and humongous as USA or China, we usually stop at certain points.
Although I personally made efforts to get past those points and make my business bigger and better, even with the limited infrastructure we have here.

devmio: What excites you most about the future of AI, beyond just the art generation aspects?

Muhammadreza Haghiri: AI is way more than the generative field we know and love. I wrote a lot of AI apps way before Mann-E and Vecentor. Such as ALPNR (Automated License Plate Number Recognition) proof-of-concept for Iranian license plates, American and Persian sign language translators, OSS Persian OCR, etc.
But in this new advanced field, I see a lot of potentials. Specially with these new methods such as function calling, we easily can do a lot of stuff such as making personal/home assistants, AI powered handhelds, etc.

Updates on Mann-E

devmio: Since our last conversation, what kind of updates and upgrades for Mann-E have you been working on?

Muhammadreza Haghiri: Mann-E is now having a new model (no SD anymore, but heavily influenced by SD), generating better images and we’re getting closer to Midjourney. To be honest, in eyes of most of our users, our outputs were much better than Dall-E 3 and Midjourney.
We have one more rival to fight (according to the feedback from users) and that is Ideogram. One thing we’ve done is that we’ve added an LLM improvement system for user prompts!

devmio: How does Mann-E handle complex or nuanced prompts compared to other AI models?
Are there any plans to incorporate user feedback into the training process to improve Mann-E’s generation accuracy?

Muhammadreza Haghiri: As I said in the previous answer, we now have an LLM in the middle of user and model (you have to check its checkbox by the way) and it takes your prompt, processes it, gives it to the model and boom, you have results even better that Midjourney!

P.S: I mention Midjourney a lot, since most of Iranian investors expected us to be exactly like current version of midjourney when even SD 1.5 was a new thing, this is why Midjourney became our benchmark and biggest rival at the same time!

Explore Generative AI Innovation

Generative AI and LLMs

Learn more

Questions about vecentor:

devmio: Can you please tell our readers more about the model underneath vecentor?

Muhammadreza Haghiri: It’s more like a combination of models or a pipeline of models. It uses an image generation model (like Mann-E’s model), then a pattern recognition model (or a vision model if you mind) and then, a code generation model generates the resulting SVG code.

This is the best way of creating SVG’s using AI, specially complex SVG’s like what we have on our platform!

devmio: Why did you choose a mixture of Mistral and Stable Diffusion?

Muhammadreza Haghiri: The code generation is done by Mistral (a finetuned version), but image generation and pattern recognition aren’t exactly done by SD.
Although at the time of our initial talks, we were still using SD, but we just switched to Mann-E’s proprietary models and trained a vector style on top of that.
Then, we just moved to OpenAI’s vision models in order to get the information about the image and the patterns.
At the end, we use our LLM in order to create the SVG code.
It’s a fun and complex task of generation of SVG images!

devmio: How does Vecentor’s approach to SVG generation differ from traditional image generation methods (like pixel-based models)?

Muhammadreza Haghiri: As I mentioned, SVG generation is being treated as code generation because vector images are more like guidelines of how lines and dots are drawn and colored on the user’s screen. Also there are some information of scales and the scales aren’t literal (hence the name “scalable”).
So we can claim that we achieved code generation in our company and it opens the doors for us to make new products for developers and people who need to code.

devmio: What are the advantages and limitations of using SVGs for image creation compared to other formats?

Muhammadreza Haghiri: For a lot of applications such as desktop publications or web development, SVG’s are better choice.
They can be easily modified and their quality will be the same. This is why SVG’s matter. The limitations on the other hand are that you just can’t expect a photo realistic image be a good SVG, since they’re made very very geometric.

devmio: Can you elaborate on specific applications where Vecentor’s SVG generation would be particularly beneficial (e.g., web design, animation, data visualization)?

Muhammadreza Haghiri: Of course. Our initial target market was for frontend developers and UI/UX designers. But it can also be spread to the other industries and professions as well. 

The Future of AI Art Generation

devmio: With the rise of AI art generators, how do you see the role of human artists evolving?

Muhammadreza Haghiri: Unlike what a lot of people think, humans are always ahead of machines. Although an intelligent machine is not without its own dangers, but we still can be far ahead of what a machine can do. Human artists will evolve and will become better of course, and we can take a page from their books and make better intelligent machines!

devmio: Do you foresee any ethical considerations specific to AI-generated art, such as copyright or plagiarism concerns?

Muhammadreza Haghiri: This is a valid concern and debate. Artists want to protect their rights and we also want more data. I guess the best way of getting rid of copyright disasters is that not being like OpenAI and if we use copyrighted material, we pay the owners of the art!
This is why both Mann-E and Vecentor are trained of AI generated and also royalty free material.

devmio: What potential applications do you see for AI art generation beyond creative endeavors?

Muhammadreza Haghiri: AI image, video and music generation is a tool for marketers in my opinion. You will have a world to create without any concerns about copyrights and what’s better than this? I personally think this is the future in those areas.
Also, I personally look at AI art as a form of entertainment. We used to listen to the music other people made, nowadays we can produce the music ourselves just by typing what we have in our mind!

Personal Future and Projects

devmio: Are you currently planning new projects or would you like to continue working on your existing projects?

Muhammadreza Haghiri: Yes. I’m planning for some projects, specially in Hardware and Code Generation areas. I guess they’ll be surprises for next quarters

devmio: Are there any areas in the field of AI or ML that you would like to explore further in the near future?

Muhammadreza Haghiri: I like the hardware and OS integrations. Something like a self operating computer or similar stuff. Also, I like to see more AI usage in our day to day lives.

devmio: Thank you very much for taking the time to answer our questions.

The post Art and creativity with AI appeared first on ML Conference.

Katleen Gabriels: My name is Katleen Gabriels, I am an Assistant Professor in Ethics and Philosophy of Technology at the University of Maastricht in the Netherlands, but I was born and raised in Belgium. I studied linguistics, literature, and philosophy. My research career started as an avatar in Second Life and the social virtual world. Back then I was a master student in moral philosophy and I was really intrigued by this social virtual world that promised that you could be whoever you want to be. 

That became the research of my master thesis and evolved into a PhD project which was on the ontological and moral status of virtual worlds. Since then, all my research revolves around the relation between morality and new technologies. In my current research, I look at the mutual shaping of morality and technology. 

Some years ago, I held a chair at the Faculty of Engineering Sciences in Brussels and I gave lectures to engineering and mathematics students and I’ve also worked at the Technical University of Eindhoven.

devmio: Where exactly do philosophy and AI overlap?

Katleen Gabriels: That’s a very good but also very broad question. What is really important is that an engineer does not just make functional decisions, but also decisions that have a moral impact. Whenever you talk to engineers, they very often want to make the world a better place through their technology. The idea that things can be designed for the better already has moral implications.

Way too often, people believe in the stereotype that technology is neutral. We have many examples around us today, and I think machine learning is a very good one, that a technology’s impact is highly dependent on design choices. For example, the data set and the quality of the data: If you train your algorithms with just even numbers, it will not know what an uneven number is. But there are older examples that have nothing to do with AI or computer technology. For instance, a revolving door does not include people who need a walking cane or a wheelchair.

In my talks, I always share a video of an automatic soap dispenser that does not recognize black people’s hands to show why it is so important to take into consideration a broad variety of end users. Morality and technology are not separate domains. Each technological object is human-made and humans are moral beings and therefore make moral decisions. 

Also, the philosophy of the mind is very much dealing with questions concerning intelligence, but with breakthroughs in generative AI like DALL-E, also, with what is creativity. Another important question that we’re constantly debating with new evolutions in technology is where the boundary between humans and machines is. Can we be replaced by a machine and to what extent?

Explore Generative AI Innovation

Generative AI and LLMs

Learn more

devmio: In your book “Conscientious AI: Machines Learning Morals”, you write a lot about design as a moral choice. How can engineers or developers make good moral choices in their design?

Katleen Gabriels: It’s not only about moral choices, but also about making choices that have ethical impact. My most practical hands-on answer would be that education for future engineers and developers should focus much more on these conceptual and philosophical aspects. Very often, engineers or developers are indeed thinking about values, but it’s difficult to operationalize them, especially in a business context where it’s often about “act now, apologize later”. Today we see a lot of attempts of collaboration between philosophers and developers, but that is very often just a theoretical idea.

First and foremost, it’s about awareness that design choices are not just neutral choices that developers make. We have had many designers with regrets about their designs years later. Chris Wetherell is a nice example: He designed the retweet button and initially thought that the effects of it would only be positive because it can increase how much the voices of minorities are heard. And that’s true in a way, but of course, it has also contributed to fake news to polarization.

Often, people tend to underestimate how complex ethics is. I exaggerate a little bit, but very often when teaching engineers, they have a very binary approach to things. There are always some students who want to make a decision tree out of ethical decisions. But often values clash with each other, so you need to find a trade-off. You need to incorporate the messiness of stakeholders’ voices, you need time for reflection, debate, and good arguments. That complexity of ethics cannot be transferred into a decision tree. 

If we really want to think about better and more ethical technology, we have to reserve a lot of time for these discussions. I know that when working for a highly commercial company, there is not a lot of time reserved for this.

devmio: What is your take on biases in training data? Is it something that we can get rid of? Can we know all possible biases?

Katleen Gabriels: We should be aware of the dynamics of society, our norms, and our values. They’re not static. Ideas and opinions, for example, about in vitro fertilization have changed tremendously over time, as well as our relation with animal rights, women’s rights, awareness for minorities, sustainability, and so on. It’s really important to realize that whatever machine you’re training, you must always keep it updated with how society evolves, within certain limits, of course. 

With biases, it’s important to be aware of your own blind spots and biases. That’s a very tricky one. ChatGPT, for example, is still being designed by white men and this also affects some of the design decisions. OpenAI has often been criticized for being naive and overly idealistic, which might be because the designers do not usually have to deal with the kind of problems they may produce. They do not have to deal with hate speech online because they have a very high societal status, a good job, a good degree, and so on.

devmio: In the case of ChatGPT, training the model is also problematic. In what way?

Katleen Gabriels: There’s a lot of issues with ChatGPT. Not just with the technology itself, but things revolving around it. You might already have read that a lot of the labeling and filtering of the data has been outsourced, for instance, to clickworkers in Africa. This is highly problematic. Sustainability is also a big issue because of the enormous amounts of power that the servers and GPUs require. 

Another issue with ChatGPT has to do with copyright. There have already been very good articles about the arrogance of Big Tech because their technology is very much based on the creative works of other people. We should not just be very critical about the interaction with ChatGPT, but also about the broader context of how these models have been trained, who the company and the people behind it are, what their arguments and values are, and so on. This also makes the ethical analysis much more complex.

The paradox is that on the Internet, with all our interactions, we become very transparent for Big Tech companies, but they in turn remain very opaque about their decisions. I’ve also been amazed and but annoyed about how a lot of people dealt with the open letter demanding a six-month ban on AI development. People didn’t look critically at people like Elon Musk signing it and then announcing the start of a new AI company to compete with OpenAI.

This letter focuses on existential threats and yet completely ignores the political and economic situation of Big Tech today. 

devmio: In your book, you wrote that language still represents an enormous challenge for AI. The book was published in 2020 – before ChatGPT’s advent. Do you still hold that belief today?

Katleen Gabriels: That is one of the parts that I will revise and extend significantly in the new edition. Even though the results are amazing in terms of language and spelling, ChatGPT still is not magic. One of the challenges of language is that it’s context specific and that’s still a problem for algorithms, which has not been solved with ChatGPT. It’s still a calculation, a prediction.

The breakthrough in NLP and LLMs indeed came sooner than I would have expected, but some of the major challenges are not being solved. 

devmio: Language plays a big role in how we think and how we argue and reason. How far do you think we are from artificial general intelligence? In your book, you wrote that it might be entirely possible, that consciousness might be an emergent property of our physiology and therefore not achievable outside of the human body. Is AGI even achievable?

Katleen Gabriels: Consciousness is a very tricky one. For AGI, first of all, from a semantic point of view, we need to know what intelligence is. That in itself is a very philosophical and multidimensional question because intelligence is not just about being good in mathematics. The term is very broad. There is also emotional and different kinds of intelligence, for instance. 

We could take a look at the term superintelligence, as the Swedish philosopher Nick Bostrom defines it: Superintelligence means that a computer is much better than a human being and each facet of intelligence, including emotional intelligence. We’re very far away from that. It also has to do with bodily intelligence. It’s one thing to make a good calculation, but it’s another thing to teach a robot to become a good waiter and balance glasses filled with champagne through a crowd. 

AGI or strong AI means a form of consciousness or self-consciousness and includes the very difficult concept of free will and being accountable for your actions. I don’t see this happening. 

The concept of AGI is often coupled with the fear of the singularity, which is basically a threshold: The final thing we as humans do, is develop a very smart computer and then we are done for as we cannot compete with these computers. Ray Kurzweil predicted that this is going to happen in 2045. But depending on the definition of superintelligence and the definition of singularity, I don’t believe that 2045 will be the time when this happens. Very few people actually believe that.

devmio: We regularly talk to our expert Christoph Henkelmann. He raised an interesting point about AGI. If we are able to build a self-conscious AI, we have a responsibility to that being and cannot just treat it as a simple machine.

Katleen Gabriels: I’m not the only person who made the joke, but maybe the true Turing Test is that if a machine gains self-consciousness and commits suicide, maybe that is a sign of true intelligence. If you look at the history of science fiction, people have been really intrigued by all these questions and in a way, it very much fits the quote that “to philosophize is to learn how to die.”

I can relate that quote to this, especially the singularity is all about overcoming death and becoming immortal. In a way, we could make sense of our lives if we create something that outlives us, maybe even permanently. It might make our lives worth living. 

At the academic conferences that I attend, the consensus seems to be that the singularity is bullshit, the existential threat is not that big of a deal. There are big problems and very real threats in the future regarding AI, such as drones and warfare. But a number of impactful people only tell us about those existential threats. 

devmio: We recently talked to Matthias Uhl who worked on a study about ChatGPT as a moral advisor. His study concluded that people do take moral advice from a LLM, even though it cannot give it. Is that something you are concerned with?

Katleen Gabriels: I am familiar with the study and if I remember correctly, they required a five minute attention span of their participants. So in a way, they have a big data set but very little has been studied. If you want to ask the question of to what extent would you accept moral advice from a machine, then you really need a much more in-depth inquiry. 

In a way, this is also not new. The study even echoes some of the same ideas from the 1970s with ELIZA. ELIZA was something like an early chatbot and its founder, Joseph Weizenbaum, was shocked when he found out that people anthropomorphized it. He knew what it was capable of and in his book “Computer Power and Human Reason: From Judgment to Calculation” he recalls anecdotes where his secretary asked him to leave the room so she could interact with ELIZA in private. People were also contemplating to what extent ELIZA could replace human therapists. In a way, this says more about human stupidity than about artificial intelligence. 

In order to have a much better understanding of how people would take or not take moral advice from a chatbot, you need a very intense study and not a very short questionnaire.

devmio:  It also shows that people long for answers, right? That we want clear and concise answers to complex questions.

Katleen Gabriels: Of course, people long for a manual. If we were given a manual by birth, people would use it. It’s also about moral disengagement, it’s about delegating or distributing responsibility. But you don’t need this study to conclude that.

It’s not directly related, but it’s also a common problem on dating apps. People are being tricked into talking to chatbots. Usually, the longer you talk to a chatbot, the more obvious it might become, so there might be a lot of projection and wishful thinking. See also the media equation study. We simply tend to treat technology as human beings.

devmio: We use technology to get closer to ourselves, to get a better understanding of ourselves. Would you agree?

Katleen Gabriels: I teach a course about AI and there’s always students saying, “This is not a course about AI, this is a course about us!” because it’s so much about what intelligence is, where the boundary between humans and machines is, and so on. 

This would also be an interesting study for the future of people who believe in a fatal singularity in the future. What does it say about them and what they think of us humans?

devmio: Thank you for your answers!

The post Building Ethical AI: A Guide for Developers on Avoiding Bias and Designing Responsible Systems appeared first on ML Conference.

DataBricks is known as the Data Lake House. This is a combination of a data warehouse and data lake. This article will take a closer look at what this means in practice and how you can start your first experiments with DataBricks.{.preface}

You should know that the DataBricks platform is a spin-off of the Apache Spark project. As with many open source projects, the idea behind it was to combine open source technology with quality of life improvements.

DataBricks in particular obviously focuses on ease of use and a flat learning curve. Developers should resist the temptation to use an inexpensive, turnkey product instead of a  technically innovative system, especially for projects with a short lifespan.

Commissioning DataBricks

DataBricks is currently used exclusively in resources or implementations from cloud providers. At the time of writing this, the company at least supports the “Big Three”. Interestingly, in the [FAQ] seen in **Figure 1**, they explicitly admit that they don’t currently provide the option of locally hosting the DataBricks system.

Fig. 1: If you want to host DataBricks locally, you’re out of luck.{.caption}

Interestingly, DataBricks has a close relationship with all three cloud providers. In many cases, you don’t have to pay separate AWS or cloud costs when purchasing a commercial DataBricks product. Instead, payment is made directly to DataBricks and the provider settles the costs.

For newcomers, there is the DataBricks Community Edition, a light version provided in collaboration with Amazon AWS. It’s completely free to use, but only allows 15 GB of data volume and is limited in terms of some convenience functions, scheduling (and the REST API). But this function should be enough for our first attempts.

So let’s call up the [DataBricks Community Edition log-in page] in the browser of our choice. After clicking on the sign-up link, DataBricks takes you to the fully-fledged log-in portal, where you can register for a free 14-day trial of the platform’s full version. In order to use the Community Edition, you must first fully complete the registration process.

In the second step, be sure not to choose a cloud provider in the window shown in **Figure 2**. Instead, click the Get started with Community Edition link at the bottom to continue the registration process for the Community Edition.

Fig. 2: Care is needed when activating the Community Edition.{.caption}

In the next step, you need to solve a captcha to identify yourself as a human user. The confirmation message seen in **Figure 3** is divided between the commercial and community edition. Don’t get anxious about the reference to the free trial phase.

Fig. 3: Community Edition users also see this message.{.caption}

Entering a valid e-mail address is especially important. DataBricks will send a confirmation email. Clicking the link in the email lets you set a password. Then you’ll find yourself in the product’s start interface, [which can be activated later here](https://community.cloud.databricks.com/).

MYRIAD OF TOOLS & FRAMEWORKS

Tools, APIs & Frameworks

Learn more

Working through the Quickstart notebook

In many respects, commercial companies are interested in flattening the learning curve for potential customers. This can be seen in DataBrick’s guide. The Quickstart tutorial section is prominently placed on the homepage, offering the Start Tutorial link.

Click it to command the web interface to change mode. Your efforts will be rewarded with a user interface similar to several Python notebook systems.

The visual similarities are no coincidence. DataBricks relies on the IPython engine in the background and is more or less compatible with standalone product versions.

Creating the cluster is especially important here. Let me explain. The developer creates the intelligence needed to complete the machine learning task in the notebooks.

But the actual execution of this intelligence requires computing power that normally far exceeds the available computing resources behind Schlomo Normaldevveloper’s browser window. Interestingly, DataBricks’ clusters are available in two versions. The all-purpose class is a classic cloud VM that (manually started and/or scheduled) is also available to a user rotation for collaboratively finishing battle tasks.

System number two is the job cluster. This is a dedicated cluster created for a batch task. It is automatically terminated after a successful or failed job processing. It’s important to note that the administrator isn’t able to keep a job cluster alive after the batch process finishes.

Be that as it may, in the next step, we place our mouse pointer on the far left to expand the menu. DataBricks offers two different operating modes by default.

We want to choose Data Science and Engineering. In the next step, open the Compute menu. Here, we can manage the computing power sources in our account.

Activate the All-Purpose-Compute tab and click the Create Compute option to make a new cluster element. You can freely choose a name. I opted for SUSTest1.

It’s important that several Runtime versions are available. In the following, we opt for the 7.3 LTS option (Scala 2.12, Spark 3.0.1).

As free Community Edition users, we don’t have the option of choosing different cluster hardware sizes. Our system only ever has 15 GB of memory and deactivates after two hours of inactivity.

So, all you need to do to start the configuration process is click the Create Cluster button. Then, click the compute element again to switch to the overview table. This lists all of your account’s compute resources side-by-side.

Generating the compute resources will take some time. To the far left of the table, as seen in **Figure 4**, there is a rotating circle symbol to show that our cluster is in progress.

Fig. 4: If the circle is rotating, the cluster isn’t ready for combat yet.{.caption}

The start process can take up to five minutes. Once the work is done, a green tick symbol will appear, as seen in **Figure 5**. As a free version user, you cannot assume that your cluster is running ad perpetuum. If you notice strange behavior in the DataBricks, it makes sense to check the cluster status.

Fig. 5: The green tick mean it’s ready for action.{.caption}

Once our work is done, we can return to the notebook. The Connect option is available in the top right-hand corner. Click it and select the cluster to establish a connection. Then click 

the Run All icon next to it to instruct all commands in the notebook to execute. In the following, the system will execute commands in individual cells in real-time, as seen in **Figure 5**. Be sure to scroll down and view the results.

Fig. 6: The environment provides real-time information about operations performed

Focus on the cell.{.caption}

Due to the architectural decision to build DataBricks as a whole on IPython notebooks, we  must deliver the commands to be executed in the form of notebooks. Interestingly, the notebook as a whole can be kept in one programming language, while individual command cells can offer other languages. A foreign-language command element is created by clicking the respective language bubble, as shown in **Figure 7**.

Fig. 7: DataBricks allows the use of insular languages.{.caption}

Using the menu option File | Export | HTML, the DataBricks notebook can also be exported as an HTML file after its commands are successfully processed. The majority of the mark-up is lost, but the resulting file presents the results in a way that’s easier for management to understand and digest.

Alternatively, you can click the blue Publish button to generate a globally valid link that lets any user view the fully-fledged notebook. By default, these links stay valid for six months. Please note that publishing a new version invalidates all existing links.

Commercial version owners can also run their notebooks regularly like a cron job with the scheduling option. The user interface in **Figure 8** is used for this. Other job scheduling system users will feel right at home. However, be aware that this function requires a job cluster, which isn’t included and cannot be created in the free Community Edition at the time of writing this.

Fig. 8: DataBricks in scheduling mode.{.caption}

Last but not least, you can also stop the cluster using the menu at the top right. This is only a courtesy to the company for the Community Edition. But, it’s highly recommended for commercial use since it reduces overall costs.

Different data tables for optimizing performance

One of NoSQL databases’ basic characteristics is that in many cases, they soften the ACID criteria. The lower consistency quality is usually offset by a greatly reduced database administration effort. Sometimes, this results in impressive performance increases compared to a classic relational database. When working with DataBricks, we deal with a group of different table types that differ in terms of performance and data storage type.

The most important difference concerns external tables and managed tables. A managed table lives entirely in the DataBricks cluster. The development team understands this to mean that the database server handles management of the actual information and the provision of metadata and access features.

There’s also the unmanaged or external table. This table represents a kind of “wrapper” around an external data source. Using this design pattern is recommended if you frequently use sample databases or information already available elsewhere in the system in an accessible form.

Since our sample from DataBricks is based on a diamond information set, using external tables is recommended. Redundant duplication of resources will only waste memory space in our cluster, without bringing any significant benefits here.

However, a careful look at the instructions created in the example notebook shows two different procedures. The first table is created with the following snippet:

```

DROP TABLE IF EXISTS diamonds;
CREATE TABLE diamonds
USING csv
OPTIONS (path "/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv", header "true")
```

Besides the call to DROP TABLE, which is always needed to initialize the cluster, creating the new table uses standard SQL commands, more or less.We use _Using csv_ to tell the Runtime we want to use the CSV engine.

If you scroll further down in the example, you’ll see that the table is created again, but in a two-stage process. In the first step, there’s now a Python island in the notebook that interacts with the diamond sample information in the URL /databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv according to the following:

```
%python
diamonds = spark.read.csv("/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv", header="true", inferSchema="true")
diamonds.write.format("delta").mode("overwrite").save("/delta/diamonds")
```

The DataBricks development team provides aspiring data science experimenters with a dozen or so widely used sample datasets. These can be accessed directly from the DataBricks runtime using friendly URLs. Additional information about available data sources [can be found here](https://docs.databricks.com/dbfs/databricks-datasets.html).

In the second step, there’s a snippet of SQL code that delivers Using Delta instead of the previously used Using CSV. This instructs the DataBricks backend to animate the existing element with the Delta database engine.

```
DROP TABLE IF EXISTS diamonds;

CREATE TABLE diamonds USING DELTA LOCATION '/delta/diamonds/'
```

Delta is an open source database engine based on Apache Parquet. Its . Normally, it’s always preferable to use the Apache Spark table because it delivers better results in terms of both ACID criteria and performance, especially when large amounts of data need to be processed.

DataBricks is more – Focus on machine learning

Until now, we operated the DataBricks runtime in engineering mode. It’s optimized for the needs of ordinary data scientists who want to perform various types of analyses. But the user interface has a special mode specifically for machine learning (**Fig. 9** shows the mode switcher) that focuses on relevant functions.

Fig. 9: This option lets you change the personality of the DataBricks interface.{.caption}

In principle, the workflow in **Figure 10** is always used. Anyone implementing this workflow in an in-house application will always work with the Auto-ML working environment sooner or later. In theory, this is available from version 9.1 at the end of Runtime, but it’s only really feature-complete when at least version 10.4 LTS ML is available on the cluster. But since this is one of the USPs of the DataBricks platform, we can assume that the product is under constant further development.

It’s advised that you check if the cluster in question is running the product’s latest version. For data engineering, DataBricks also offers a dedicated tutorial in the Guide: Training section from the home screen. This makes it easier to get started. Click the Start guide option again to load the notebook for this tutorial as “to be edited”.

Fig. 10: If you want to use the ML functions in DataBricks, you should familiarize yourself with this workflow.{.caption}

Due to higher demands on the aforementioned required Data Bricks Runtime, you should switch to the Compute section and delete the previously created cluster. Then, click the Create Compute option again and delete the previously created cluster. Click the Create Compute option again and make sure to click the ML heading in the DataBricks Runtime Version field (see **Fig. 11**) in the first step.

Fig. 11: ML-capable variants of the DataBricks runtime appear in a separate section in the backend.{.caption}

Just for fun, we’ll use the latest version 12.0 ML and name the cluster “SUSTestML”. It takes some time after clicking the Create Cluster button, since the cloud resources aren’t immediately provided.

During cluster generation, we can return to the notebook to get an overview of the elements. In the first step, we see the inclusion of the following libraries, abbreviated here. They are familiar to every Python developer:

```

import mlflow
import numpy as np
import pandas as pd
import sklearn.datasets
. . .
from hyperopt import fmin, tpe, hp, SparkTrials, Trials, STATUS_OK
. . .
```

In many respects, DataBricks is based on what ML developers are familiar with from working with standard Python scripts. Some libraries naturally have optimizations to make them run more efficiently on the DataBricks hardware. In general, however, a locally functioning Python script will continue to work without any problems after being moved to the DataBricks cluster. For the actual monitoring of the learning process, Data Bricks relies on MLFlow, which is available here [6].

For this reason, the rest of the notebook is standard ML code, although it’s elegantly integrated into the user interface. For example, there is a flyout in which the application provides information about various parameters that were created during the parameterization of the model:

```
with mlflow.start_run(run_name='gradient_boost') as run:
  model = sklearn.ensemble.GradientBoostingClassifier(random_state=0)
  model.fit(X_train, y_train)
  . . .
```

It’s also interesting to note that the results of the individual optimization runs are not only displayed in the user interface. The Python code that lives in the notebook can also access them programmatically. In this way, it can perform a kind of reflection to find the most suitable parameters and/or model architectures.

In the case of the example notebook provided by DataBricks, this is illustrated in the following snippet, which applies an SQL query to the results available in the mlflow.search_runs field:

“`

best_run = mlflow.search_runs(
  order_by=['metrics.test_auc DESC', 'start_time DESC'],
  max_results=10,
).iloc[0]
print('Best Run')
print('AUC: {}'.format(best_run["metrics.test_auc"]))
print('Num Estimators: {}'.format(best_run["params.n_estimators"]))
```

 AutoML, for the second time

The duality of control via the user interface and programmatic control also continues in the case of the AutoML library mentioned above. The user interface shown in Figure 12, which allows graphical parameterization of ML runs, is probably the most common marketing argument.

Fig. 12: AutoML allows the graphical configuration of modeling{.caption}

On the other hand, there is also a programmatic API that illustrates DataBricks in the form of a group of notebooks. Here we want to use the example notebook provided here [7], which we load into a browser window in the first step. Then click on the Import Notebook button at the top right and copy the URL to the clipboard.

Next, open the menu of your DataBricks instance and select the Workspace File Users option. Next to your email address, there is a downward pointing arrow, which allows you to open a context menu. Select the import option there and then enter the URL to load the sample notebook into your DataBricks instance.

The actual body of the model couldn’t be any easier. In the first step, we mainly load test data, but we also create a schema element that informs the engine about the type or data type of the model information to be processed:

```

from pyspark.sql.types import DoubleType, StringType, StructType, StructField

schema = StructType([
  StructField("age", DoubleType(), False),
  . . .
  StructField("income", StringType(), False)
])
input_df = 
spark.read.format("csv").schema(schema).load("/databricks-datasets/adult/adult.data")
```
The actual classification run then also takes place with a single line:
```

from databricks import automl
summary = automl.classify(train_df, target_col="income", timeout_minutes=30)
```

If you want to carry out interference later, you can do this with both Pandas and Spark.

The multitool for ML professionals

Although there are hundreds of pages yet to be written about DataBricks, we’ll end our experiments with this brief overview. DataBricks is a tool that is completely focused on data scientists and machine learning experts and is not really suitable for beginners due to the very steep learning curve. Much like the infamous Squirrel Busters, DataBricks is a product that will find you when you need it.

The post Maximizing Machine Learning with Data Lakehouse and Databricks: A Guide to Enhanced AI Workflows appeared first on ML Conference.

Embeddings are essentially a technique for converting unstructured data into a structure that can be easily processed by software. They are used to transform complex data such as words, sentences, or even entire documents into a vector space, with similar elements close to each other. These vector representations allow machines to recognize and exploit nuances and relationships in the data. Which is essential for a variety of applications such as natural language processing (NLP), image recognition, and recommendation systems.

OpenAI, the company behind ChatGPT, offers models for creating embeddings for texts, among other things. At the end of January 2024, OpenAI presented new versions of these embeddings models, which are more powerful and cost-effective than their predecessors. In this article, after a brief introduction to embeddings, we’ll take a closer look at the OpenAI embeddings and the recently introduced innovations, discuss how they work, and examine how they can be used in various software development projects.

Embeddings briefly explained

Imagine you’re in a room full of people and your task is to group these people based on their personality. To do this, you could start asking questions about different personality traits. For example, you could ask how open someone is to new experiences and rate the answer on a scale from 0 to 1. Each person is then assigned a number that represents their openness.

Next, you could ask about another personality trait, such as the level of sense of duty, and again give a score between 0 and 1. Now each person has two numbers that together form a vector in a two-dimensional space. By asking more questions about different personality traits and rating them in a similar way, you can create a multidimensional vector for each person. In this vector space, people who have similar vectors can then be considered similar in terms of their personality.

In the world of artificial intelligence, we use embeddings to transform unstructured data into an n-dimensional vector space. Similarly how a person’s personality traits are represented in the vector space, each point in this vector space represents an element of the original data (such as a word or phrase) in a way that is understandable and processable by computers.

OpenAI Embeddings

OpenAI embeddings extend this basic concept. Instead of using simple features like personality traits, OpenAI models use advanced algorithms and big data to achieve a much deeper and more nuanced representation of the data. The model not only analyzes individual words, but also looks at the context in which those words are used, resulting in more accurate and meaningful vector representations.

Another important difference is that OpenAI embeddings are based on sophisticated machine learning models that can learn from a huge amount of data. This means that they can recognize subtle patterns and relationships in the data that go far beyond what could be achieved by simple scaling and dimensioning, as in the initial analogy. This leads to a significantly improved ability to recognize and exploit similarities and differences in the data.

Explore Generative AI Innovation

Generative AI and LLMs

Learn more

Individual values are not meaningful

While in the personality trait analogy, each individual value of a vector can be directly related to a specific characteristic – for example openness to new experiences or a sense of duty – this direct relationship no longer exists with OpenAI embeddings. In these embeddings, you cannot simply look at a single value of the vector in isolation and draw conclusions about specific properties of the input data. For example, a specific value in the embedding vector of a sentence cannot be used to directly deduce how friendly or not this sentence is.

The reason for this lies in the way machine learning models, especially those used to create embeddings, encode information. These models work with complex, multi-dimensional representations where the meaning of a single element (such as a word in a sentence) is determined by the interaction of many dimensions in vector space. Each aspect of the original data – be it the tone of a text, the mood of an image, or the intent behind a spoken utterance – is captured by the entire spectrum of the vector rather than by individual values within that vector.

Therefore, when working with OpenAI embeddings, it’s important to understand that the interpretation of these vectors is not intuitive or direct. You need algorithms and analysis to draw meaningful conclusions from these high-dimensional and densely coded vectors.

Comparison of vectors with cosine similarity

A central element in dealing with embeddings is measuring the similarity between different vectors. One of the most common methods for this is cosine similarity. This measure is used to determine how similar two vectors are and therefore the data they represent.

To illustrate the concept, let’s start with a simple example in two dimensions. Imagine two vectors in a plane, each represented by a point in the coordinate system. The cosine similarity between these two vectors is determined by the cosine of the angle between them. If the vectors point in the same direction, the angle between them is 0 degrees and the cosine of this angle is 1, indicating maximum similarity. If the vectors are orthogonal (i.e. the angle is 90 degrees), the cosine is 0, indicating no similarity. If they are opposite (180 degrees), the cosine is -1, indicating maximum dissimilarity.

Figure 1 -Cosine similarity

Accompanying this article is a Google Colab Python Notebook which you can use to try out many of the examples shown here. Colab, short for Colaboratory, is a free cloud service offered by Google. Colab makes it possible to write and execute Python code in the browser. It’s based on Jupyter Notebooks, a popular open-source web application that makes it possible to combine code, equations, visualizations, and text in a single document-like format. The Colab service is well suited for exploring and experimenting with the OpenAI API using Python.

In practice, especially when working with embeddings, we are dealing with n-dimensional vectors. The calculation of the cosine similarity remains conceptually the same, even if the calculation is more complex in higher dimensions. Formally, the cosine similarity of two vectors A and B in an n-dimensional space is calculated by the scalar product (dot product) of these vectors divided by the product of their lengths:

Figure 2 – Calculation of cosine similarity

The normalization of vectors plays an important role in the calculation of cosine similarity. If a vector is normalized, this means that its length (norm) is set to 1. For normalized vectors, the scalar product of two vectors is directly equal to the cosine similarity since the denominators in the formula from Figure 2 are both 1. OpenAI embeddings are normalized, which means that to calculate the similarity between two embeddings, only their scalar product needs to be calculated. This not only simplifies the calculation, but also increases efficiency when processing large quantities of embeddings.

OpenAI Embeddings API

OpenAI offers a web API for creating embeddings. The exact structure of this API, including code examples for curl, Python and Node.js, can be found in the OpenAI reference documentation.

OpenAI does not use the LLM from ChatGPT to create embeddings, but rather specialized models. They were developed specifically for the creation of embeddings and are optimized for this task. Their development was geared towards generating high-dimensional vectors that represent the input data as well as possible. In contrast, ChatGPT is primarily optimized for generating and processing text in a conversational form. The embedding models are also more efficient in terms of memory and computing requirements than more extensive language models such as ChatGPT. As a result, they are not only faster but much more cost-effective.

New embedding models from OpenAI

Until recently, OpenAI recommended the use of the text-embedding-ada-002 model for creating embeddings. This model converts text into a sequence of floating point numbers (vectors) that represent the concepts within the content. The ada v2 model generated embeddings with a size of 1536 dimensions and delivered solid performance in benchmarks such as MIRACL and MTEB, which are used to evaluate model performance in different languages and tasks.

At the end of January 2024, OpenAI presented new, improved models for embeddings:

text-embedding-3-small: A smaller, more efficient model with improved performance compared to its predecessor. It performs better in benchmarks and is significantly cheaper.
text-embedding-3-large: A larger model that is more powerful and creates embeddings with up to 3072 dimensions. It shows the best performance in the benchmarks but is slightly more expensive than ada v2.

A new function of the two new models allows developers to adjust the size of the embeddings when generating them without significantly losing their concept-representing properties. This enables flexible adaptation, especially for applications that are limited in terms of available memory and computing power.

Readers who are interested in the details of the new models can find them in the announcement on the OpenAI blog. The exact costs of the various embedding models can be found here.

Create embeddings with Python

In the following section, the use of embeddings is demonstrated using a few code examples with Python. The code examples are designed so that they can be tried out in Python Notebooks. They are also available in a similar form in the previously mentioned accompanying Google Colab notebook mentioned above.
Listing 1 shows how to create embeddings with the Python SDK from OpenAI. In addition, numpy is used to show that the embeddings generated by OpenAI are normalized.

Listing 1

from openai import OpenAI
from google.colab import userdata
import numpy as np

# Create OpenAI client
client = OpenAI(
    api_key=userdata.get('openaiKey'),
)

# Define a helper function to calculate embeddings
def get_embedding_vec(input):
  """Returns the embeddings vector for a given input"""
  return client.embeddings.create(
        input=input,
        model="text-embedding-3-large", # We use the new embeddings model here (announced end of Jan 2024)
        # dimensions=... # You could limit the number of output dimensions with the new embeddings models
    ).data[0].embedding

# Calculate the embedding vector for a sample sentence
vec = get_embedding_vec("King")
print(vec[:10])

# Calculate the magnitude of the vector. I should be 1 as
# embedding vectors from OpenAI are always normalized.
magnitude = np.linalg.norm(vec)
magnitude

Similarity analysis with embeddings

In practice, OpenAI embeddings are often used for similarity analysis of texts (e.g. searching for duplicates, finding relevant text sections in relation to a customer query, and grouping text). Embeddings are very well suited for this, as they work in a fundamentally different way to comparison methods based on characters, such as Levenshtein distance. While it measures the similarity between texts by counting the minimum number of single-character operations (insert, delete, replace) required to transform one text into another, embeddings capture the meaning and context of words or sentences. They consider the semantic and contextual relationships between words, going far beyond a simple character-based level of comparison.

As a first example, let’s look at the following three sentences (the following examples are in English, but embeddings work analogously for other languages and cross-language comparisons are also possible without any problems):

I enjoy playing soccer on weekends.
Football is my favorite sport. Playing it on weekends with friends helps me to relax.
In Austria, people often watch soccer on TV on weekends.

In the first and second sentence, two different words are used for the same topic: Soccer and football. The third sentence contains the original soccer, but it has a fundamentally different meaning from the first two sentences. If you calculate the similarity of sentence 1 to 2, you get 0.75. The similarity of sentence 1 to 3 is only 0.51. The embeddings have therefore reflected the meaning of the sentence and not the choice of words.

Here is another example that requires an understanding of the context in which words are used:
He is interested in Java programming.
He visited Java last summer.
He recently started learning Python programming.

In sentence 2, Java refers to a place, while sentences 1 and 3 have something to do with software development. The similarity of sentence 1 to 2 is 0.536, but that of 1 to 3 is 0.587. As expected, the different meaning of the word Java has an effect on the similarity.

The next example deals with the treatment of negations:
I like going to the gym.
I don’t like going to the gym.
I don’t dislike going to the gym.

Sentences 1 and 2 say the opposite, while sentence 3 expresses something similar to sentence 1. This content is reflected in the similarities of the embeddings. Sentence 1 to sentence 2 yields a cosine similarity of 0.714 while sentence 1 compared to sentence 3 yields 0.773. It is perhaps surprising that there is no major difference between the embeddings. However, it’s important to remember that all three sets are about the same topic: The question of whether you like going to the gym to work out.

The last example shows that the OpenAI embeddings models, just like ChatGPT, have built in a certain “knowledge” of concepts and contexts through training with texts about the real world.

I need to get better slicing skills to make the most of my Voron.
3D printing is a worthwhile hobby.
Can I have a slice of bread?

In order to compare these sentences in a meaningful way, it’s important to know that Voron is the name of a well-known open-source project in the field of 3D printing. It’s also important to note that slicing is a term that plays an important role in 3D printing. The third sentence also mentions slicing, but in a completely different context to sentence 1. Sentence 2 mentions neither slicing nor Voron. However, the trained knowledge enables the OpenAI Embeddings model to recognize that sentences 1 and 2 have a thematic connection, but sentence 3 means something completely different. The similarity of sentence 1 and 2 is 0.333 while the comparison of sentence 1 and 3 is only 0.263.

Similarity values are not percentages

The similarity values from the comparisons shown above are the cosine similarity of the respective embeddings. Although the cosine similarity values range from -1 to 1, with 1 being the maximum similarity and -1 the maximum dissimilarity, they are not to be interpreted directly as percentages of agreement. Instead, these values should be considered in the context of their relative comparisons. In applications such as searching text sections in a knowledge base, the cosine similarity values are used to sort the text sections in terms of their similarity to a given query. It is important to see the values in relation to each other. A higher value indicates a greater similarity, but the exact meaning of the value can only be determined by comparing it with other similarity values. This relative approach makes it possible to effectively identify and prioritize the most relevant and similar text sections.

Embeddings and RAG solutions

Embeddings play a crucial role in Retrieval Augmented Generation (RAG) solutions, an approach in artificial intelligence that combines the capabilities of information retrieval and text generation. Embeddings are used in RAG systems to retrieve relevant information from large data sets or knowledge databases. It is not necessary for these databases to have been included in the original training of the embedding models. They can be internal databases that are not available on the public Internet.
With RAG solutions, queries or input texts are converted into embeddings. The cosine similarity to the existing document embeddings in the database is then calculated to identify the most relevant text sections from the database. This retrieved information is then used by a text generation model such as ChatGPT to generate contextually relevant responses or content.

Vector databases play a central role in the functioning of RAG systems. They are designed to efficiently store, index and query high-dimensional vectors. In the context of RAG solutions and similar systems, vector databases serve as storage for the embeddings of documents or pieces of data that originate from a large amount of information. When a user makes a request, this request is first transformed into an embedding vector. The vector database is then used to quickly find the vectors that correspond most closely to this query vector – i.e. those documents or pieces of information that have the highest similarity. This process of quickly finding similar vectors in large data sets is known as Nearest Neighbor Search.

Challenge: Splitting documents

A detailed explanation of how RAG solutions work is beyond the scope of this article. However, the explanations regarding embeddings are hopefully helpful for getting started with further research on the topic of RAGs.

However, one specific point should be pointed out at the end of this article: A particular and often underestimated challenge in the development of RAG systems that go beyond Hello World prototypes is the splitting of longer texts. Splitting is necessary because the OpenAI embeddings models are limited to just over 8,000 tokens. One token corresponds to approximately 4 characters in the English language (see also).

It’s not easy finding a good strategy for splitting documents. Naive approaches such as splitting after a certain number of characters can lead to the context of text sections being lost or distorted. Anaphoric links are a typical example of this. The following two sentences are an example:

VX-2000 requires regular lubrication to maintain its smooth operation.
The machine requires the DX97 oil, as specified in the maintenance section of this manual.

The machine in the second sentence is an anaphoric link to the first sentence. If the text were to be split up after the first sentence, the essential context would be lost, namely that the DX97 oil is necessary for the VX-2000 machine.

There are various approaches to solving this problem, which will not be discussed here to keep this article concise. However, it is essential for developers of such software systems to be aware of the problem and understand how splitting large texts affects embeddings.

Summary

Embeddings play a fundamental role in the modern AI landscape, especially in the field of natural language processing. By transforming complex, unstructured data into high-dimensional vector spaces, embeddings enable in-depth understanding and efficient processing of information. They form the basis for advanced technologies such as RAG systems and facilitate tasks such as information retrieval, context analysis, and data-driven decision-making.

OpenAI’s latest innovations in the field of embeddings, introduced at the end of January 2024, mark a significant advance in this technology. With the introduction of the new text-embedding-3-small and text-embedding-3-large models, OpenAI now offers more powerful and cost-efficient options for developers. These models not only show improved performance in standardized benchmarks, but also offer the ability to find the right balance between performance and memory requirements on a project-specific basis through customizable embedding sizes.

Embeddings are a key component in the development of intelligent systems that aim to achieve useful processing of speech information.

Links and Literature:

1. https://colab.research.google.com/gist/rstropek/f3d4521ed9831ae5305a10df84a42ecc/embeddings.ipynb
2. https://platform.openai.com/docs/api-reference/embeddings/create
3. https://openai.com/blog/new-embedding-models-and-api-updates
4. https://openai.com/pricing
5. https://platform.openai.com/tokenizer

The post OpenAI Embeddings appeared first on ML Conference.

1. Geospatial Analysis: Accurate address matching forms the foundation of geospatial analysis, allowing organizations to make informed decisions about locations, market trends, and resource allocation across various industries like retail and media.
2. Real Estate Data Management: In the real estate industry, precise address matching facilitates property valuation, market analysis, and portfolio management.
3. Logistics and Navigation: Efficient routing and delivery depend on accurate address matching.
4. Compliance and Regulation: Many regulatory requirements mandate precise address data, such as tax reporting and census data collection.

Cherre is the leading real estate data management company, we specialize in accurate address matching for the second use case. Whether you’re an asset manager, portfolio manager, or real estate investor, a building represents the atomic unit of all financial, legal, and operating information. However, real estate data lives in many silos, which makes having a unified view of properties difficult. Address matching is an important step in breaking down data silos in real estate. By joining disparate datasets on address, we can unlock many opportunities for further portfolio analysis.

Data Silos in Real Estate

Real estate data usually fall into the following categories: public, third party, and internal. Public data is collected by governmental agencies and made available publicly, such as land registers. The quality of public data is generally not spectacular and the data update frequency is usually delayed, but it provides the most comprehensive coverage geographically. Don’t be surprised if addresses from public data sources are misaligned and misspelled.

Third party data usually come from data vendors, whose business models focus on extracting information as datasets and monetizing those datasets. These datasets usually have good data quality and are much more timely, but limited in geographical coverage. Addresses from data vendors are usually fairly clean compared to public data, but may not be the same address designation across different vendors. For large commercial buildings with multiple entrances and addresses, this creates an additional layer of complexity.

Lastly, internal data is information that is collected by the information technology (I.T.) systems of property owners and asset managers. These can incorporate various functions, from leasing to financial reporting, and are often set up to represent the business organizational structures and functions. Depending on the governance standards, and data practices, the quality of these datasets can vary and data coverage only encompasses the properties in the owner’s portfolio. Addresses in these systems can vary widely, some systems are designated at the unit-level, while others designate the entire property. These systems also may not standardize addresses inherently, which makes it difficult to match property records across multiple systems.

With all these variations in data quality, coverage, and address formats, we can see the need for having standardized addresses to do basic property-level analysis.

Address Matching Using the Parse-Clean-Match Strategy

In order to match records across multiple datasets, the address parse-clean-match strategy works very well regardless of region. By breaking down addresses into their constituent pieces, we have many more options for associating properties with each other. Many of the approaches for this strategy use simple natural language processing (NLP) techniques.

NEW & PRACTICAL ENDEAVORS FOR ML

Machine Learning Principles

Learn more

Address Parsing

Before we can associate addresses with each other, we must first parse the address. Address parsing is the process of breaking down each address string into its constituent components. Components in addresses will vary by country.

In the United States and Canada, addresses are generally formatted as the following:

{street_number} {street_name}

{city}, {state_or_province} {postal_code}

{country}

In the United Kingdom, addresses are formatted very similarly as in the U.S. and Canada, with an additional optional locality designation:

{building_number} {street_name}

{locality (optional)}

{city_or_town}

{postal_code}

{country}

French addresses vary slightly from U.K. addresses with the order of postal code and city:

{building_number} {street_name}

{postal_code} {city}

{country}

German addresses take the changes in French addresses and then swap the order of street name and building number:

{street_name} {building_number} {postal_code} {city} {country}

Despite the slight variations across countries’ address formats, addresses generally have the same components, which makes this an easily digestible NLP problem. We can break down the process into the following steps:

1. Tokenization: Split the address into its constituent words. This step segments the address into manageable units.
2. Named Entity Recognition (NER): Identify entities within the address, such as street numbers, street names, cities, postal codes, and countries. This involves training or using pre-trained NER models to label the relevant parts of the address.
3. Sequence Labeling: Use sequence labeling techniques to tag each token with its corresponding entity

Let’s demonstrate address parsing with a sample Python code snippet using the spaCy library. SpaCy is an open-source software library containing many neural network models for NLP functions. SpaCy supports models across 23 different languages and allows for data scientists to train custom models for their own datasets. We will demonstrate address parsing using one of SpaCy’s out-of-the-box models for the address of a historical landmark: David Bowie’s Berlin apartment.

import spacy

# Load the NER spaCy model
model = spacy.load("en_core_web_sm")

# Address to be parsed
address = "Hauptstraße 155, 10827 Berlin"

# Tokenize and run NER
doc = model(address)

# Extract address components
street_number = ""
street_name = ""
city = ""
state = ""
postal_code = ""

for token in doc:
    if token.ent_type_ == "GPE":  # Geopolitical Entity (City)
        city = token.text
    elif token.ent_type_ == "LOC":  # Location (State/Province)
        state = token.text
    elif token.ent_type_ == "DATE":  # Postal Code
        postal_code = token.text
    else:
        if token.is_digit:
            street_number = token.text
        else:
            street_name += token.text + " "

# Print the parsed address components
print("Street Number:", street_number)
print("Street Name:", street_name)
print("City:", city)
print("State:", state)
print("Postal Code:", postal_code)

Now that we have a parsed address, we can now clean each address component.

Address Cleaning

Address cleaning is the process of converting parsed address components into a consistent and uniform format. This is particularly important for any public data with misspelled, misformatted, or mistyped addresses. We want to have addresses follow a consistent structure and notation, which will make further data processing much easier.

To standardize addresses, we need to standardize each component, and how the components are joined. This usually entails a lot of string manipulation. There are many open source libraries (such as libpostal) and APIs that can automate this step, but we will demonstrate the basic premise using simple regular expressions in Python.

import pandas as pd
import re

# Sample dataset with tagged address components
data = {
    'Street Name': ['Hauptstraße', 'Schloß Nymphenburg', 'Mozartweg'],
    'Building Number': ['155', '1A', '78'],
    'Postal Code': ['10827', '80638', '54321'],
    'City': ['Berlin', ' München', 'Hamburg'],
}

df = pd.DataFrame(data)

# Functions with typical necessary steps for each address component

# We uppercase all text for easier matching in the next step

def standardize_street_name(street_name):

    # Remove special characters and abbreviations, uppercase names
    standardized_name = re.sub(r'[^\w\s]', '', street_name)
    return standardized_name.upper()

def standardize_building_number(building_number):
    # Remove any non-alphanumeric characters (although exceptions exist)
    standardized_number = re.sub(r'\W', '', building_number)
    return standardized_number

def standardize_postal_code(postal_code):

    # Make sure we have consistent formatting (i.e. leading zeros)
    return postal_code.zfill(5)

def standardize_city(city):

    # Upper case the city, normalize spacing between words
    return ' '.join(word.upper() for word in city.split())

# Apply standardization functions to our DataFrame
df['Street Name'] = df['Street Name'].apply(standardize_street_name)
df['Building Number'] = df['Building Number'].apply(standardize_building_number)
df['Postal Code'] = df['Postal Code'].apply(standardize_postal_code)
df['City'] = df['City'].apply(standardize_city)

# Finally create a standardized full address (without commas)

df[‘Full Address’] = df['Street Name'] + ' ' + df['Building Number'] + ' ' + df['Postal Code'] + ' ' + df['City']

Address Matching

Now that our addresses are standardized into a consistent format, we can finally match addresses from one dataset to address in another dataset. Address matching involves identifying and associating similar or identical addresses from different datasets. When two full addresses match exactly, we can easily associate the two together through a direct string match.

When addresses don’t match, we will need to apply fuzzy matching on each address component. Below is an example of how to do fuzzy matching on one of the standardized address components for street names. We can apply the same logic to city and state as well.

from fuzzywuzzy import fuzz

# Sample list of street names from another dataset
street_addresses = [
    "Hauptstraße",
    "Schlossallee",
    "Mozartweg",
    "Bergstraße",
    "Wilhelmstraße",
    "Goetheplatz",
]

# Target address component (we are using street name)
target_street_name = "Hauptstrasse " # Note the different spelling and space 

# Similarity threshold

# Increase this number if too many false positives

# Decrease this number if not enough matches

threshold = 80

# Perform fuzzy matching
matches = []

for address in street_addresses:
    similarity_score = fuzz.partial_ratio(address, target_street_name)
    if similarity_score >= threshold:
        matches.append((address, similarity_score))

matches.sort(key=lambda x: x[1], reverse=True)

# Display matched street name
print("Target Street Name:", target_street_name)
print("Matched Street Names:")
for match in matches:
    print(f"{match[0]} (Similarity: {match[1]}%)")

Up to here, we have solved the problem for properties with the same address identifiers. But what about the large commercial buildings with multiple addresses?

Other Geospatial Identifiers

Addresses are not the only geospatial identifiers in the world of real estate. An address typically refers to the location of a structure or property, often denoting a street name and house number.  There are actually four other geographic identifiers in real estate:

1. A “lot” represents a portion of land designated for specific use or ownership.
2. A “parcel” extends this notion to a legally defined piece of land with boundaries, often associated with property ownership and taxation.
3. A “building” encompasses the physical structures erected on these parcels, ranging from residential homes to commercial complexes.
4. A “unit” is a sub-division within a building, typically used in multi-unit complexes or condominiums. These can be commercial complexes (like office buildings) or residential complexes (like apartments).

What this means is that we actually have multiple ways of identifying real estate objects, depending on the specific persona and use case. For example, leasing agents focus on the units within a building for tenants, while asset managers optimize for the financial performance of entire buildings. The nuances of these details are also codified in many real estate software systems (found in internal data), in the databases of governments (found in public data), and across databases of data vendors (found in third party data). In public data, we often encounter lots and parcels. In vendor data, we often find addresses (with or without units). In real estate enterprise resource planning systems, we often find buildings, addresses, units, and everything else in between.

In the case of large commercial properties with multiple addresses, we need to associate various addresses with each physical building. In this case, we can use geocoding and point-in-polygon searches.

Geocoding Addresses

Geocoding is the process of converting addresses into geographic coordinates. The most common form is latitude and longitude. European address geocoding requires a robust understanding of local address formats, postal codes, and administrative regions. Luckily, we have already standardized our addresses into an easily geocodable format.

Many commercial APIs exist for geocoding addresses in bulk, but we will demonstrate geocoding using a popular Python library, Geopy, to geocode addresses.

from geopy.geocoders import Nominatim

geolocator = Nominatim(user_agent="my_geocoder")
location = geolocator.geocode("1 Canada Square, London")
print(location.latitude, location.longitude)

Now that we’ve converted our addresses into latitude and longitude, we can use point-in-polygon searches to associate addresses with buildings.

Point-in-Polygon Search

A point-in-polygon search is a technique to determine if a point is located within the boundaries of a given polygon.

The “point” in a point-in-polygon search refers to a specific geographical location defined by its latitude and longitude coordinates. We have already obtained our points by geocoding our addresses.

The “polygon” is a closed geometric shape with three or more sides, which is usually characterized by a set of vertices (points) connected by edges, forming a closed loop. Building polygons can be downloaded from open source sites like OpenStreetMap or from specific data vendors. The quality and detail of the OpenStreetMap building data may vary, and the accuracy of the point-in-polygon search depends on the precision of the building geometries.

While the concept seems complex, the code for creating this lookup is quite simple. We demonstrate a simplified example using our previous example of 1 Canada Square in London.

import json
from shapely.geometry import shape, Point

# Load the GeoJSON data
with open('building_data.geojson') as geojson_file:
    building_data = json.load(geojson_file)

# Latitude and Longitude of 1 Canada Square in Canary Wharf
lat, lon = 51.5049, 0.0195

# Create a Point geometry for 1 Canada Square
point_1_canada = Point(lon, lat)

# See if point is within any of the polygons
for feature in building_data['features']:
    building_geometry = shape(feature['geometry'])

    if point_1_canada.within(building_geometry):
        print(f"Point is within this building polygon: {feature}")
        break
else:
    print("Point is not within any building polygon in the dataset.")

Using this technique, we can properly identify all addresses associated with this property.

Summary

Addresses in real life are confusing because they are the physical manifestation of many disparate decisions in city planning throughout the centuries-long life of a city. But using addresses to match across different datasets doesn’t have to be confusing.

Using some basic NLP and geocoding techniques, we can easily associate property-level records across various datasets from different systems. Only through breaking down data silos can we have more holistic views of property behaviors in real estate.

Author Biography

Alyce Ge is data scientist at Cherre, the industry-leading real estate data management and analytics platform. Prior to joining Cherre, Alyce held data science and analytics roles for a variety of technology companies focusing on real estate and business intelligence solutions. Alyce is a Google Cloud-certified machine learning engineer, Google Cloud-certified data engineer, and Triplebyte certified data scientist. She earned her Bachelor of Science in Applied Mathematics from Columbia University in New York.

The post Address Matching with NLP in Python appeared first on ML Conference.