The Exchanges

Q Okay. Any, uh, lessons learned so far? I mean, as, as you said, you guys come from a very deep background in, uh, in AI and AI research. It's a whole different thing to build a whole like data center facility. How have you gone about it? And, uh, what, what are some things that, that surprised you and any lessons so far?

A As most new experiences as a founder, I relied on the knowledge of others, and so I was lucky to have a very, a few seasoned HPC experts and a lot of cloud software experts as well to build that solution. For me personally, and it's one of the things I love about my position at Mistral is that I get to discover so many new things and so many new problems I hadn't thought possible. Having to learn to like all of the different parts of building a data center, all of the different trades that you have to coordinate, uh, all of the potential, um, synchronization, uh, between all of the different trades. I mean, it's a huge building. It involves hundreds of people working on it. You have this, then when you, uh, stand up the thing, uh, you have to question what works. You have to filter through the blades that are faulty. It's just an entire new area of work where I get to see, um, Experts in their field go through things and try to explain to me what their daily work is. It's always fascinating to see, um, an expert in this field, like do something that you don't know how to do. I think the logistics of it, uh, and the timelines are also quite different from what I'm usually, um, dealing with in software and research. For new capacity to, uh, be built, you have to plan around, uh, Having energy available, you have to plan for the, uh, space to be available and on time. And so it's …

AI assessment note: “I think the logistics of it, uh, and the timelines are also quite different”

Q And who does the, all of this you have evolved towards an FDE model?

A So we have indeed a large, uh, FDE section. It's, it's a mix of software and And, uh, FDs, and we split our FDs into what we called, um, AI engineers and applied scientists. Um, and so, uh, applied scientists will tend to use the tools that we've just, uh, uh, talked about. So fine-tuning, continued pre-training, and the likes, where AI engineers will focus more on adaptation to the enterprise environment and figuring out what workflows to automate and all of this. They work with the customers to make sure that the use cases are indeed providing values and going to production, but it's also a fantastic way for us to understand what matters in an enterprise context and be faster at building the right platform.

AI assessment note: “It's, it's a mix of software and And, uh, FDs”

Q As I'm sure you've seen, there was, uh, for the last few weeks in startup and venture circles, there's been this whole, um, idea of the context graph as an infrastructure that made the rounds. Is that something that you think about or a layer that would, uh, basically, uh, enable one to know how the agents made a decision and, uh, how those decisions relate to one another?

A I've seen this indeed, and I think there are Two, uh, levels to that discussion. The part that you mentioned at the end where, uh, it's interesting to know how an agent came to. So in that discussion, when, when we talk about understanding how an agent came to a decision or an action, the game is really to understand how a human, uh, agent really made this decision. It's understanding how an enterprise does what it does. And it's certainly interesting. Uh, what keeps me up at night and what I really want to solve first is just the basic idea of gathering a workable enterprise context. Right now, uh, with, uh, any model, uh, and with a lot of effort, you will be able to get some connections to tools and you will ask your questions and your agent will do a bunch of things. It will realize that, oh, by doing Five API calls and three joins. I can probably get, uh, what Timothy asked immediately. What should happen is that, um, all of that, uh, discovery and all of that intelligence should be stored somewhere to be reused. It's not really how things happen. It's just basic knowledge about what the infrastructure of the company is. So knowing where the tables are, what they contain, how they're joined. So all of this, um, is, Compute that should be amortized, basically. And to me, it's really the entire game with the context engine, as we call it internally, is to be in a setup where…

AI assessment note: “I've seen this indeed, and I think there are Two, uh, levels to that discussion.”

Q Where does the edge fit in all of this?

A There are a few reasons to go edge. First, there are some regions where it's more convenient to be able to work without internet, and there are also a lot of capabilities that don't necessarily require a huge model. So if you just need something that goes voice to action, On any device, uh, today with, uh, typically the voxel models that we develop, this is doable. Again, an area where The more focused your use case is, the smaller you can make the model through fine-tuning or through just distillation in an even smaller architecture. I think voice to action is going to be a big use case. I think it will simplify a lot the current stacks for these types of things. There is also some privacy things where you could imagine all of the context consolidation stays On your personal device. And for most things, uh, you can deal with a small model, uh, that answers a lot of your questions. And then you potentially can get, uh, what goes out to, uh, another like cloud-based models. I myself take the train a lot. Uh, I like having coding assistance, uh, having, uh, DevStroll run on my laptop while I could on the train is, uh, comfortable despite the bad wifi.

AI assessment note: “There are a few reasons to go edge. First, there are some regions”

Q Another thing that you, uh, released reasonably recently, I believe is, uh, OCR three. What does that do? That enables you to just like, uh, scan Any, uh, any form, any document?

A Yeah. OCR is a huge use case in enterprise. Uh, a lot of our customers have, I mean, the typical example is KYC where someone will submit a form and you need to input that information in a structured way in your systems, or you need to reason about it. And so OCR, interestingly, is, uh, it, it's not the types of systems that I would have expected, uh, LLMs to really, uh, make large strides on. The visual reasoning and the visual understanding has gotten so good that it's, it's just an easier way to process things. Uh, in my mind, you have any sort of inputs, um, and you can get the, the data that you care about. As I mentioned, when you build agents, you have a different type of inputs for the tasks that you're trying to solve. Documents and visual informations are just a very, very frequent kind of kind of input. Uh, sometimes it's a lot cheaper Uh, to use a small OCR model to just get the text that you care about and then potentially post-process it or deal with it with another system than to run it through a large, uh, multimodal model that will, uh, basically do the same thing but at a higher cost.

AI assessment note: “Yeah. OCR is a huge use case in enterprise.”

Q So let's go into, uh, the enterprise reality of all of this. Um, so if I'm an enterprise or if I'm a sovereign and I want to deploy a Mr. Open source model, what is it that I do these days with everything that you, that you've built?

A The way we work with, um, enterprise, I mean, as you mentioned, like we have a few of our models that are open source and Apache and All of our clients are welcome to use them, uh, as they need. What we have seen in terms of success is that given the current stack, it still requires, um, a lot of expertise, uh, to manage to come to, um, actual value and, um, and things that go to production, basically. The way we interact is that we usually stand up our, um, Mistral AI Studio, which is our platform, and we can deploy all of our stack on the client's choice of deployment methods, so it can be on-prem, it can be on their VPC, it can be on in several places. The reason we do this is that it lets clients build where their data is, and without having to shuffle things around, which, as I've learned as a CTO, is something that you don't want to do ever. Because it asks, it raises a lot of questions, and it's quite a stressful thing to do. So once this is deployed, we then work with the business units to understand where their pain points are. Sometimes it's knowledge management, and I think it's the most well-known use case from the output, from the outside of the enterprise world. But it's also around automating core workflows for the enterprise. It's, you know, some Tooling that you wouldn't expect where one thing that we've done is around code modernization, uh, where you're, you …

AI assessment note: “we can deploy all of our stack on the client's choice of deployment methods”

Q What is the current constraint, uh, that, um, you guys are facing to make, uh, Mistral four when it eventually comes out, uh, do much, much better than Mistral three? Is that a question of Mistral compute or is that a question of, of data? And, uh, in particular, are you guys doing anything around synthetic data that you can talk about?

A Definitely compute and, uh, the current deployment that we have will help, uh, as it's going to be giving us a lot more Grace Blackwell capacity than we had in the past. And so that's, uh, something that we're very excited about. And when you add a compute, you also have to add data. And so we've been hard at work, uh, making sure that our, uh, data mixtures are, uh, as high quality as ever and growing in size. But as you mentioned, One of the ways to do this is through synthetic, synthetic data. In terms of, um, where we use synthetic data the most, I think a lot of the interesting work that's happening is for the post-training part where we can, um, build environments, uh, that look similar to, uh, an enterprise and then, uh, try to, uh, synthetically create queries that are hard and that will require multiple hops. And so all of this work, um, is in addition to the coding work, the reasoning work is really what makes the final model able to perform, uh, in the various, uh, environments that we work in. So before it was about, uh, accruing world knowledge and the, uh, web helps a lot with this. Now it's more and more about acquiring know-how. Uh, and for this, uh, it's really about Um, trying to find what our customers are trying to do, trying to replicate it inside of our training environment, and, ah, let the, the model run, basically.

AI assessment note: “Definitely compute and, uh, the current deployment that we have will help”

Q So just to set it up in the last year, you guys raised A 1.7 billion euros, uh, Series C led by ASML at an 11.7 billion post-money valuation. You launch a bunch of models, which we're going to talk about. Is the big vision behind all of this that enterprises and sovereign states are going to need their own AI infrastructure and Mistrol is going to be the provider?

A So the big vision has been evolving. And as you stated, we started Uh, as a company that built models, uh, because with Arthur and Guillaume, this was what we knew how to do at the start. The premise on which we built Mitchell AI was immediately solving for enterprise needs. Uh, and we started with open weights model after this, uh, and working with enterprise, we realized, uh, the need for basically the rest of the stack. So we built, uh, the serving platform because infrastructure was needed. Um, and then All of the tooling around it, uh, was also something that we saw was missing more than the tooling. It also requires a lot of work and expertise still to get deep into, uh, an enterprise workflows and really help that transformation. And so we built that, uh, FDE function and more recently, uh, with missile compute, uh, we're going a bit lower, uh, in the stack as well. So we've done all of this, uh, because it was required for enterprise success. Uh, while still continuing, uh, on our models journey. All of this stack, uh, being modular is really important to us as it gives full control to, uh, enterprise and our clients as to which part of the stack they decide to, uh, own and control, which is maybe more involved or that they decide to have serverless or basically this modularity that we like.

AI assessment note: “The premise on which we built Mitchell AI was immediately solving for enterprise needs.”

Q You mentioned post-training, and that's one of the key topics of the last 12 months in particular, this evolution of LLMs into systems with both pre-training and post-training and a lot of reinforcement learning. Where do you guys fall in that spectrum? Are you pushing a lot of reinforcement learning? Do you believe that pre-training has still room to grow? How do you think about it?

A Yeah, everything still has room to grow. What I'm interested in as the CTO is really how you make, uh, all of the steps of the pipeline, uh, work well together. Now everyone can, uh, develop most efficiently. Um, typically what happens in, uh, post-training Is that you will have a team that's working on improving code. You will have another team that's improving, um, different, uh, enterprise, uh, behaviors. You will have another team that's, uh, improving on, uh, instruction following. Uh, and so all of this, uh, at some point has to come together because customers aren't happy if you require them to deploy five different, uh, models to get their job done. There is really an internal engine and capability around making all of these work stream come together in the way that you expect. That is super interesting to build. And so, but yeah, uh, internally we're building and improving all of the parts of the stack. I think the post training is very rich because it also touches all of the new use cases of LLMs. And I think it's been very exciting To see just all of the, the new use case that pop up every day. Anytime someone on Twitter finds a new exciting things that they've done, then suddenly, you know, you've got to make it this proof of concept into potentially a base capability on which your model will perform well. And that's, uh, potentially an entire stream of work and you…

AI assessment note: “everything still has room to grow. What I'm interested in as the CTO”

Q the kind of the banger, uh, use cases in the enterprise? Let's assume like all agents work, uh, in, you know, in a workflow kind of a way that you described, uh, based on either your, uh, industry watch or, or more specifically talking to your customers, uh, What is it that is going to generate a amazing ROI beyond coding, which is pretty established at this, at this stage?

A Yeah, there are several dimensions to this. Coding is an obvious one. And, um, to me to get the full, um, ROI of coding, you need customization, uh, because a lot of ROI is unlocked, uh, on like sprawling code bases that are completely impossible to know for, uh, for something that's been trained on the web. Uh, if you've got, uh, an enterprise that's been building its own, like domain specific languages for years. You'll need some customization for an agent to come in and be competent, uh, in that respect. Um, so coding is definitely a big one. Um, if everything, uh, comes true as I hope, I think there is still a huge jump in how we accelerate knowledge worker. Um, and I believe the magical experience of, uh, you go to your chat assistant, it's connected to your system and you can ask it Anything, uh, about the enterprise just hasn't realized yet. And it's really obvious, uh, when you see the kind of queries that people are making, expecting them to just work. And to me, who's building the system, it, it feels like magic. Like if you need to somehow send an email to three people and coordinate a meeting and also like gather data from some BI system, it's just something that requires, um, a lot more plumbing and capabilities that we have today. Um, so that's going to be a huge lift. And I think the last one, which is maybe closer to my heart, is really when we start to customiz…

AI assessment note: “I think there is still a huge jump in how we accelerate knowledge worker.”

Q And from a team building perspective, how have you gone about it? Uh, the, the, the three of you, the three co-founders of a deep background in, uh, in AI, um, are you this day focused mostly on building like an FDA team or are you still, uh, building this large kind of like research lab effort? And how do you, uh, think about the right ratio?

A We are growing, uh, all of our teams, both, uh, research, uh, FDs, uh, product engineering, uh, infrastructure for compute, and all of the teams have their own, uh, challenges in how you build and what order you, uh, recruit people in. It's been important to me, um, at the start to, I mean, to me and, uh, and Guillaume and Arthur, we Both, like the three of us were, uh, good AI practitioners. So we knew how to train models and we knew how to code. And so we started with people like us to get to the models trained the fastest. Um, but that's doesn't work as you scale. Uh, you, it is critical to build the right, uh, infrastructure, uh, for research. And so this takes different skill sets. Uh, and it's something that we've been, uh, building over the years as well. Uh, and it's, Fascinating as someone who used to do research at a smaller scale to see the kind of systems that are involved and the gains that you can have at scale. In terms of engineering, it's kind of the same story, really, where you start with a team that's broad in its knowledge and self-sufficient and can iterate fast, and then more and more you bring in experts or people that are, that have seen larger scale and will tell you like, well, This won't work in six months, and so we should fix that now. So it's been super interesting growing the company and seeing all of the successive things that break at each scal…

AI assessment note: “We are growing, uh, all of our teams, both, uh, research, uh, FDs”

Q It's a good reminder that like you guys have, I've done all of this and the company was started in, right? Yeah. June, 23, right? If I recall.

A Yeah. And so for most of our clients, uh, we, we started working with them recently. The tooling, uh, for everyone is still in its infancy. And so I hope that the tooling will stabilize, uh, and I hope that we will have true value. True value to me is really, okay, we've gone through that first phase of building connections and now employees of that enterprise are able to use everything that we've built. Right now, I think we're in a phase where we build siloed things. Because we're scared of data going through walls and everything. And so to me, the real success is when you're confident enough to give all of that control back to the company's employees at large, and they start really building on it.

AI assessment note: “Yeah. And so for most of our clients, uh, we, we started working”

Q What is the ultimate goal of the model effort? I mean, clearly you guys are a frontier AI lab, but, um, are you trying to create the best models and, and solve AGI? Or are you trying to be the best open source model compared to the Chinese labs or, you know, whatever open source eventually comes out of the, uh, US? What is it that you're trying to do?

A We're trying to get the best, uh, models that we can and the model that's most useful for, uh, the use cases that we cover, uh, in enterprise. And so typically with the rise of, uh, agentic, uh, behavior, one thing that's very important is how you deal with, uh, various contexts, how you deal with various, um, documents, uh, being added to the input. And so having the capabilities to do architecture iterations, Really trying new things in terms of model training is critical. Um, so we're pushing the boundaries of what the current models can do with, uh, the compute capacity that we have, but we're also trying to focus on the things that are, is most annoying, uh, in our deployments today. And so one of the consideration that has been solved with a few harness, uh, tricks is the Context of those agentic systems. So it's visible typically in vibe coding, but it's, um, definitely, uh, applicable to a lot of other use cases where through all of the tool calls, you'll have to, uh, consolidate, uh, and summarize the context to be able to fit everything and, uh, have the model focus on the right parts. To me, this is just an artifact of the current architectures. Uh, we're trying to fit Uh, things in a linear context windows where essentially the questions that we're asking aren't really necessarily all linear. Um, and so we rely today on the file system for this, and that I think tha…

AI assessment note: “We're trying to get the best models that we can and the model that's most useful”

Q Okay. This is super interesting. Obviously the, the, the key question about agents these days, especially when they are combined into workflows is the question of, uh, autonomy. How do you guys think about it? How autonomous are those agents, uh, in, in your deployments?

A I don't know if it's the way I think about it. To me, the better question usually is how much you trust the agents, and there are a few dimensions around this. What worries me when building those kind of workflows is that typically if you want the value to accrue, and if you want to build faster and faster, the more workflows that you build, what you will want to do is reuse assets and make them reusable by others. As soon as you do this with agents, you then start to ask the question, well, this agent has access to some data that is privileged, but maybe this other agent is publishing it to something that's public. You might have governance concerns where some agent is acting on something very critical, and you don't know necessarily that the data that it got has been approved or something like this. It's really a new way to develop where Uh, the parts of your workflows have to be trusted. Each of them to be trusted requires, uh, quite a lot of tooling, uh, and quite a lot of observability, uh, to get confidence and to basically enable this at scale in an enterprise. So the question that you're asking about autonomy, to me, this is something that I see happening when I vibe code. Sure, like longer running tasks and making and improving on this is going to be critical and we're, uh, working on it daily, but Today, the problems that we're solving on the software side of things a…

AI assessment note: “To me, the better question usually is how much you trust the agents”