The Exchanges

Q to run this with some chronology. I spoke to so many of your advisors, investors, uh, users even. Um, and I want to start with actually the first model, you know, Michelle seven B being one of the most popular released, you know, a while ago now. Why do you think it was so popular? What do you think you did so right? And what did you learn from that?

A I think it, it served two purposes. So the first was to, uh, show that, um, there was a lot of, uh, slack in compressing models. And so from a scientific perspective, it was a good finding and, and a good learning for, for, from the community. Um, it also filled the gap, uh, in the efficiency to performance, uh, to the space of models, uh, where there was definitely something missing. And seven B is the size, uh, that allows to run Efficiently a model on your Macbook or on your smartphone. Um, and, and we made it sufficiently smart so that it was still useful. So there was already seven B models before, but they weren't good enough to do interesting applications. And so by, by targeting this specific space, uh, we talked to the developers immediately because developers like the casual developers running on a Like gaming GPU or on its MacBook. So it, it created a lot of curiosity and adoption because it was a missing spot in the, in the performance to efficiency space.

AI assessment note: “created a lot of curiosity and adoption because it was a missing spot”

Q Arthur at Mistral, I'm so intrigued to hear your thoughts. How do you think about the focus on just model improvement, where value accrues, and ultimately are the foundation models commoditizing in themselves?

A It's two opposing directions. So the first is that the models are getting better and better. So it means that creating a verticalized application, as long as you have the data for it and a good understanding of the use case you're facing, is going to be easier and easier if you have access to the tools that facilitate it. Uh, so that's the first aspect, which would make me think that, uh, the application layer is going to grow thinner and thinner. But then there's also the fact that the models are, are getting, uh, cheaper and cheaper because we managed to compress them because we make a lot of improvement on their efficiency. And so that means that effectively this plus the competitive pressure there is on the model layer means that, uh, the price around the model, the dollar per intelligence unit, let's say is definitely going to reduce. So there's these two aspects of growing ability, Compressed price, which on one side says that the application layer is going to grow thin, and on the other side says that the model part is going to grow thin. So for us, the, the approach that we are taking is that the model part is still going to be big enough, and that we need to build this platform on top of that, because that's where we are going to enable all of the vertical applications that will be interesting for humanity.

AI assessment note: “the model part is still going to be big enough, and that we need”

Q Do you think open source is ready for enterprise? Or do you think enterprises are ready for open source and do they care about it enough?

A It depends on the enterprises, but some have been early adopters and are using a lot of mistral models into in production. So for sure they're ready enough. Uh, no, in order to like bring them to the next level of putting things into, uh, like a large scale production, et cetera. I think they're still lacking some products around like managing, uh, correctly load balancing, uh, customizing the models because you can do it with DIY solutions, but if you want to make it robust enough and scalable enough, it's actually not easy. And if you want to actually increase the quality of the models, the custom models, uh, the recipe are, are a bit hard to, to set. So the most technical savvy enterprises are definitely ready for it. And there's a few, there's Actually many, uh, uh, use cases that are in production using, uh, uh, using open source models. No, in order to widen the adoption, there's definitely some tooling to be brought to the market.

AI assessment note: “the most technical savvy enterprises are definitely ready for it”

Q How do you think about that positioning and brand? Cause there are other players who are much more direct and saying, Hey, you know, we'll, you know, we're going to dominate a lot of different verticals and kind of be afraid. How do you think about that enabler to vertical applications or not in that positioning?

A We are not a verticalized company. Uh, We started Mistral to, uh, bring value to developers and to bring freedom to developers. So, uh, when we started, there was basically, uh, one API out there, uh, soon two. Um, and the field of generative AI was starting to look like it would be very centralized around a couple of players. And we took this platform approach where the model that we're making and the technology that we are making, we are allowing developers to own it, to modify it. And so bringing freedom to developers and AI application makers is I think the best way in, in, um, distributing generative AI as widely as possible, which is our objective as a company, making AI ubiquitous, uh, bringing frontier AI into everyone's head. Uh, is the reason why we started. I think we started, we did a good job at it. We, I mean, we, we still have a lot of things ahead, but, uh, this open source part was, I believe, a good enabler for the community and made people realize that they could build very interesting technology by modifying the models themselves instead of depending on the APIs of a couple of providers.

AI assessment note: “We are not a verticalized company.”

Q to run this with some chronology. I spoke to so many of your advisors, investors, uh, users even. Um, and I want to start with actually the first model, you know, Michelle seven B being one of the most popular released, you know, a while ago now. Why do you think it was so popular? What do you think you did so right? And what did you learn from that?

A I think it, it served two purposes. So the first was to, uh, show that, um, there was a lot of, uh, slack in compressing models. And so from a scientific perspective, it was a good finding and, and a good learning for, for, from the community. Um, it also filled the gap, uh, in the efficiency to performance, uh, to the space of models, uh, where there was definitely something missing. And seven B is the size, uh, that allows to run Efficiently a model on your Macbook or on your smartphone. Um, and, and we made it sufficiently smart so that it was still useful. So there was already seven B models before, but they weren't good enough to do interesting applications. And so by, by targeting this specific space, uh, we talked to the developers immediately because developers like the casual developers running on a Like gaming GPU or on its MacBook. So it, it created a lot of curiosity and adoption because it was a missing spot in the, in the performance to efficiency space.

AI assessment note: “it created a lot of curiosity and adoption because it was a missing spot”

Q So I asked Sam Altman this question. What is the end state for the model landscape? Most people say, ah, it'll become commoditized. And actually there'll be 12 players and it'll be a race to the bottom. What is the end state for models in your mind? And how do you think about the commoditization question?

A I think the end state is, uh, to have a more developed, uh, well, uh, more features on developer platforms that allows to do customization that allows to Make low latency models that serve a certain purpose that allows to evaluate them and to improve them over time. And so the model is only a, like a tiny part. I mean, it's a central part, but it remains a tiny part of an application. And what you want to do across time and when you deploy an application, uh, that you're exposed to users, you want to ensure that it works, ensure that is, that its latency reduces over time, ensure that its quality increases over time. And so I think that's the, the end state is models are effectively going to be A starting point for any AI application developer. Uh, they need to be surrounded by tools, by, um, lifecycle management, uh, platform basically, and that's the one thing that, uh, we started to build. Like general purpose models are a bit undifferentiated, but the differentiation that you need to create for, uh, for your application comes from the data you put into it, the user feedback that you gather and the intelligence that you have to figure out what the application should be doing. And that is not commoditized at all. There's no, Recipe that, uh, allows to go from a A general purpose model to model that is super good and better than all of the others at your specific task. And I t…

AI assessment note: “general purpose models are a bit undifferentiated... that is not commoditized at all.”

Q How do you think that it's hard when you suddenly have, you know, some closed and you start building an enterprise team. For you as a founder now, how do you think about that balance between a research team and a sales team and making sure that the two cultures come together well?

A I think some important thing is to create, uh, empathy. So ensure that, uh, the science team also understand, uh, the problems that the Users are facing. It improves the science because at the end of the day, the general purpose technology we're making, Is only general purpose if you identify the use cases. So that comes back to the earlier discussion we had. So ensuring that the science team has some relatively direct exposure to the product and to the business team is actually important to make them understand what, where the model is failing and how it could be improved significantly. Um, and on the other side, the go to market team has to understand it. It's a very technical sales, uh, uh, sales motion. Cause you, you're selling not the product, but you're selling something that is going to pull out the product. So you need to tell the customer, uh, how, how these things should be used to actually make something that brings value to the business. Uh, and that only goes through strong enablement of the go to market team. So it's, I think it's a challenge. Uh, they don't operate on the same scale. Uh, the science team has cycles of several months. The go to market team, uh, goes faster, uh, as shorter cycles, let's say. Um, uh, but I think so far, uh, we've managed to recruit Go to market people that have some technical interest and, uh, and technical people that have some bu…

AI assessment note: “recruit Go to market people that have some technical interest and technical people”

Q What chance do you think that Europe has in AI? I know it sounds deterministic and defeatist, and so you might be, oh, fuck Harry, shut up. But it's like, what chance do you think Europe has in AI? And what, Does it take for us to stand up as a serious AI industry with Europe?

A I guess the chance it has is that it's a revolution and it's changing the way we do software. And so as every revolution, it opens a lot of opportunity for, uh, for new actors. And there's no reason why there shouldn't be an actor that is, that was created in Europe, uh, and that, that could grow pretty fast. And that's the mission that we, we gave ourselves. Uh, we have the talent, uh, capital, Can cross, uh, oceans without, without too much problem. Uh, we have the market. The market is more fragmented than in the US for sure. The ecosystem, the digital native ecosystem is definitely smaller, but it exists and it's growing. So there's, uh, there's local opportunity for business development. Uh, on the talent side, uh, we can hire 23, 24 years old people that we can onboard in four months and they operate as well as, uh, any, uh, software engineer in the Valley. So it's, uh, it's, people are quite talented here. And so we, if we manage to keep them and to, uh, To convince them not to go to the US, we have a lot of opportunities.

AI assessment note: “it opens a lot of opportunity for, uh, for new actors.”

Q Dude, what do AI developers care about? Everyone kind of gets on Twitter and goes, oh, did you see X's performance this week is better than Y's performance last week? What do they care about? Efficiency, scale, cost? What drives that usage and decision making?

A They care about, um, cost for sure. Uh, they care about customization, uh, being able to modify the models at will. And on that aspect, I think we are only scratching the surface of what can be done. Like the fine tuning aspect that has been, uh, like the go-to solution is probably a little too low level from What, uh, we should be doing. They care about being able to deploy anywhere. So they operate in a certain space, in a certain cloud. Uh, they, uh, might be operating on-prem. They might have some edge devices to deploy to. Um, and they want to be able to put that technology there. And so they also care about portability, which in turn offer data control. Uh, usually LLMs AI is becomes very useful when you connect it to knowledge bases or to, uh, anything that is related to certain business. In that respect, it becomes a very sensitive part of your application because it sees everything. It sees all of the data you have. And so. Uh, enterprises, for instance, do care about ensuring that the proprietary data they have, um, is accessed in something that they can, they can completely secure. Uh, and that's the reason why we, we deployed our platform on Azure and AWS, for instance, that is bringing the security layer that they need.

AI assessment note: “They care about, um, cost for sure. Uh, they care about customization”

Q Do you think open source is ready for enterprise? Or do you think enterprises are ready for open source and do they care about it enough?

A It depends on the enterprises, but some have been early adopters and are using a lot of mistral models into in production. So for sure they're ready enough. Uh, no, in order to like bring them to the next level of putting things into, uh, like a large scale production, et cetera. I think they're still lacking some products around like managing, uh, correctly load balancing, uh, customizing the models because you can do it with DIY solutions, but if you want to make it robust enough and scalable enough, it's actually not easy. And if you want to actually increase the quality of the models, the custom models, uh, the recipe are, are a bit hard to, to set. So the most technical savvy enterprises are definitely ready for it. And there's a few, there's Actually many, uh, uh, use cases that are in production using, uh, uh, using open source models. No, in order to widen the adoption, there's definitely some tooling to be brought to the market.

AI assessment note: “It depends on the enterprises, but some have been early adopters”

Q I spoke to Sarah Gou before the show, and she said the core question that I think is, you know, bluntly with the focus on efficiency and the efficiency frontier does scale matter?

A Well, scale matters in the sense that, uh, if you spend more training compute, you can make the models more compressed. So you do need to have some compute to compress models. Uh, no, scale isn't the only recipe, the only, uh, ingredients to the recipe you need to scale, but you also need to have proper data. Otherwise you reach, uh, some data quality limit. You need to have proper techniques for training. Uh, you need to, yeah, figure out a few. There's, I mean, people call it compute multiplayer, I guess. Uh, how do you actually make some efficiency gain that are not costing you compute because computing is expensive. And so the, one of the things that we do at Mistral is to try and harvest these, uh, compute multipliers.

AI assessment note: “scale matters in the sense that, uh, if you spend more training compute”

Q How do you think about that positioning and brand? Cause there are other players who are much more direct and saying, Hey, you know, we'll, you know, we're going to dominate a lot of different verticals and kind of be afraid. How do you think about that enabler to vertical applications or not in that positioning?

A We are not a verticalized company. Uh, We started Mistral to, uh, bring value to developers and to bring freedom to developers. So, uh, when we started, there was basically, uh, one API out there, uh, soon two. Um, and the field of generative AI was starting to look like it would be very centralized around a couple of players. And we took this platform approach where the model that we're making and the technology that we are making, we are allowing developers to own it, to modify it. And so bringing freedom to developers and AI application makers is I think the best way in, in, um, distributing generative AI as widely as possible, which is our objective as a company, making AI ubiquitous, uh, bringing frontier AI into everyone's head. Uh, is the reason why we started. I think we started, we did a good job at it. We, I mean, we, we still have a lot of things ahead, but, uh, this open source part was, I believe, a good enabler for the community and made people realize that they could build very interesting technology by modifying the models themselves instead of depending on the APIs of a couple of providers.

AI assessment note: “We are not a verticalized company. Uh, We started Mistral to, uh, bring value to developers”

Q Is there actually much of a barrier to creating a foundational model company today? I know that's a really broad, stupid question in many respects, but you have so many different players now and new ones popping up every day. Is the barrier just reducing day by day?

A I don't think it is, uh, to create the, I mean, the, to be relevant in that space, uh, is a very hard topic. Uh, it's, uh, you need to be, uh, dominating on the cost efficiency, on the efficiency, uh, performance by the front. And there isn't, there's only a few companies that, that are, that are currently well positioned. So you can try and do something, but, but if it's not relevant, if it's strictly dominated by another model or another technology, then you have a problem. And it's, uh, there's a few barriers that are pretty hard to, to face that you need to, uh, you need to accrue sufficient, well, to raise sufficient capital to have enough compute and be relevant. You need to have, uh, people that knows how to train models, which is still a scarce resource. And then you need to have a good brand, uh, because as you've said, it's highly competitive and this is not something that comes out of thin air. Uh, so. Yeah, I think there's still a lot of defensibility on the market, although there is a lot of noise, which is different.

AI assessment note: “I don't think it is... there's still a lot of defensibility on the market”

Q How do you think that it's hard when you suddenly have, you know, some closed and you start building an enterprise team. For you as a founder now, how do you think about that balance between a research team and a sales team and making sure that the two cultures come together well?

A I think some important thing is to create, uh, empathy. So ensure that, uh, the science team also understand, uh, the problems that the Users are facing. It improves the science because at the end of the day, the general purpose technology we're making, Is only general purpose if you identify the use cases. So that comes back to the earlier discussion we had. So ensuring that the science team has some relatively direct exposure to the product and to the business team is actually important to make them understand what, where the model is failing and how it could be improved significantly. Um, and on the other side, the go to market team has to understand it. It's a very technical sales, uh, uh, sales motion. Cause you, you're selling not the product, but you're selling something that is going to pull out the product. So you need to tell the customer, uh, how, how these things should be used to actually make something that brings value to the business. Uh, and that only goes through strong enablement of the go to market team. So it's, I think it's a challenge. Uh, they don't operate on the same scale. Uh, the science team has cycles of several months. The go to market team, uh, goes faster, uh, as shorter cycles, let's say. Um, uh, but I think so far, uh, we've managed to recruit Go to market people that have some technical interest and, uh, and technical people that have some bu…

AI assessment note: “recruit Go to market people that have some technical interest and, uh, and technical people”

Q Obviously every enterprise today is sitting in a boardroom going, what's our AI strategy? What do you advise them and what questions should they be asking?

A Start thinking about how they are going to change all of their products, uh, using AI as a premise, uh, using the existence of, uh, like very clever agents, because you can build very clever agents today, um, assuming that presence and, and And working backwards to understand the consequence in terms of organization. Uh, so to be not thinking about generative AI as a way to As a way of increasing productivity in a world processing, but rather as a way to change completely the way you operate your core business, which usually involve taking models and customizing them pretty heavily to create the differentiation that you will need in like five years time when everybody will have adopted the technology in its core business.

AI assessment note: “rather as a way to change completely the way you operate your core business”

Q is you're in France, I'm in London. We both know that European enterprises do not move very fast. Most do not even have Slack. My concern is that we drastically overestimate adoption in the near future, and maybe underestimated in the 10 year, 20 year future. Do you think that's the case here, and do you worry about the lethargy of a lot of enterprises, especially in Europe, in adoption?

A I mean, it's a general, uh, phenomenon in, uh, in tech that you always, uh, over, overestimate the speed, but underestimate the impact. I think it's probably occurring today. It's slightly different, uh, in the sense that there's some, uh, executive support for, uh, pushing generative AI solutions, even in Europe. So there's some delay compared to the U S market for sure. Uh, but it's not, I wouldn't say it's, uh, it's, it's very significant, uh, It's one year maximum in terms of delay. The challenge here is that it's a technology that is, that can take many forms. And so trying to focus on some specific thing that you can bring to the market that have AI in it, uh, is a prioritization challenge. And so you need to be very strategic around that. And it's, I don't think this is super easy for enterprises. Generally, it will become easier once they try out like a off the shelf solutions a bit more. Uh, once they realize that there are some developer platforms that allows to do it, uh, without hiring, uh, very expensive and hard to find AI scientists in house. Uh, and so we expect that this is. Going to accelerate in the, in the coming years.

AI assessment note: “I think it's probably occurring today... some delay compared to the U S market”

Q What chance do you think that Europe has in AI? I know it sounds deterministic and defeatist, and so you might be, oh, fuck Harry, shut up. But it's like, what chance do you think Europe has in AI? And what, Does it take for us to stand up as a serious AI industry with Europe?

A I guess the chance it has is that it's a revolution and it's changing the way we do software. And so as every revolution, it opens a lot of opportunity for, uh, for new actors. And there's no reason why there shouldn't be an actor that is, that was created in Europe, uh, and that, that could grow pretty fast. And that's the mission that we, we gave ourselves. Uh, we have the talent, uh, capital, Can cross, uh, oceans without, without too much problem. Uh, we have the market. The market is more fragmented than in the US for sure. The ecosystem, the digital native ecosystem is definitely smaller, but it exists and it's growing. So there's, uh, there's local opportunity for business development. Uh, on the talent side, uh, we can hire 23, 24 years old people that we can onboard in four months and they operate as well as, uh, any, uh, software engineer in the Valley. So it's, uh, it's, people are quite talented here. And so we, if we manage to keep them and to, uh, To convince them not to go to the US, we have a lot of opportunities.

AI assessment note: “I guess the chance it has is that it's a revolution”

Q When you were raising money, was it very different speaking to European investors versus US investors?

A I guess in the seed round, no, it wasn't that different because it was a seed round, uh, for the series A, which was a bigger round. It was, uh, uh, we, I mean, uh, European funds were unstructured to, uh, do the kind of deal that we were proposing. Uh, so we didn't even have a lot of conversation cause they, they just couldn't get their head around the investment that needed to be made, uh, as well as we were a pre-revenue company. Yeah. I think what is lacking and it's related to, to the ecosystem part, uh, in, in, in Europe, our, our growth funds, uh, that are able to, uh, take huge bets, uh, with lots of conviction and that in some should improve over time, especially if, uh, If we manage to, uh, to use a European wealth and channel it more into that growth funds, uh, than it is today.

AI assessment note: “in the seed round, no, it wasn't that different... for the series A”

Q I spoke to Sarah Gou before the show, and she said the core question that I think is, you know, bluntly with the focus on efficiency and the efficiency frontier does scale matter?

A Well, scale matters in the sense that, uh, if you spend more training compute, you can make the models more compressed. So you do need to have some compute to compress models. Uh, no, scale isn't the only recipe, the only, uh, ingredients to the recipe you need to scale, but you also need to have proper data. Otherwise you reach, uh, some data quality limit. You need to have proper techniques for training. Uh, you need to, yeah, figure out a few. There's, I mean, people call it compute multiplayer, I guess. Uh, how do you actually make some efficiency gain that are not costing you compute because computing is expensive. And so the, one of the things that we do at Mistral is to try and harvest these, uh, compute multipliers.

AI assessment note: “scale matters in the sense that, uh, if you spend more training compute”

Q Dude, what do AI developers care about? Everyone kind of gets on Twitter and goes, oh, did you see X's performance this week is better than Y's performance last week? What do they care about? Efficiency, scale, cost? What drives that usage and decision making?

A They care about, um, cost for sure. Uh, they care about customization, uh, being able to modify the models at will. And on that aspect, I think we are only scratching the surface of what can be done. Like the fine tuning aspect that has been, uh, like the go-to solution is probably a little too low level from What, uh, we should be doing. They care about being able to deploy anywhere. So they operate in a certain space, in a certain cloud. Uh, they, uh, might be operating on-prem. They might have some edge devices to deploy to. Um, and they want to be able to put that technology there. And so they also care about portability, which in turn offer data control. Uh, usually LLMs AI is becomes very useful when you connect it to knowledge bases or to, uh, anything that is related to certain business. In that respect, it becomes a very sensitive part of your application because it sees everything. It sees all of the data you have. And so. Uh, enterprises, for instance, do care about ensuring that the proprietary data they have, um, is accessed in something that they can, they can completely secure. Uh, and that's the reason why we, we deployed our platform on Azure and AWS, for instance, that is bringing the security layer that they need.

AI assessment note: “They care about, um, cost for sure. Uh, they care about customization”

Q Is there actually much of a barrier to creating a foundational model company today? I know that's a really broad, stupid question in many respects, but you have so many different players now and new ones popping up every day. Is the barrier just reducing day by day?

A I don't think it is, uh, to create the, I mean, the, to be relevant in that space, uh, is a very hard topic. Uh, it's, uh, you need to be, uh, dominating on the cost efficiency, on the efficiency, uh, performance by the front. And there isn't, there's only a few companies that, that are, that are currently well positioned. So you can try and do something, but, but if it's not relevant, if it's strictly dominated by another model or another technology, then you have a problem. And it's, uh, there's a few barriers that are pretty hard to, to face that you need to, uh, you need to accrue sufficient, well, to raise sufficient capital to have enough compute and be relevant. You need to have, uh, people that knows how to train models, which is still a scarce resource. And then you need to have a good brand, uh, because as you've said, it's highly competitive and this is not something that comes out of thin air. Uh, so. Yeah, I think there's still a lot of defensibility on the market, although there is a lot of noise, which is different.

AI assessment note: “I don't think it is, uh, to create the, I mean, the, to be relevant”

Q Sam and Brad said the other day that models just aren't actually that good any, like yet, and they need to improve a lot in quality. What are the largest constraints or bottlenecks on model quality today and what needs to change for them to improve?

A I think the data quality is, is a constraint. Uh, how do you ensure that the model, how do you leverage the entire world knowledge and ensure that the model follows a certain path toward learning more and more complex things? Uh, that's a very important part. And I think it has been a neglected part. There's obviously compute, but, uh, given the amount of data you have, uh, we have at hand, uh, compute is already running into, uh, is no longer the bottleneck. The bottleneck is more the data at that point. You should look at text to text models. Uh, and so the question is, how do you refine the data, and how do you feed very high quality data to the model itself, uh, in order to improve it over time? And I think in, in that setting, it becomes a bit You, you, one bottleneck that is associated to, uh, bringing better model performance is the question of how do you evaluate these performances? You need to have very good evaluation that targets very specific topics. Like you want the model to be good at, uh, helping diagnosis in, in, in mid, uh, in hospital, but in French. And oftentimes you're a bit out of domain compared to the data you have. And that's where you, you should identify a gap and you should try and fill it out. Uh, so the pushing the model capabilities become, uh, like also a question of mapping where they're failing and figuring out ways of improving it. So for ins…

AI assessment note: “The bottleneck is more the data at that point. You should look at text”

Q is you're in France, I'm in London. We both know that European enterprises do not move very fast. Most do not even have Slack. My concern is that we drastically overestimate adoption in the near future, and maybe underestimated in the 10 year, 20 year future. Do you think that's the case here, and do you worry about the lethargy of a lot of enterprises, especially in Europe, in adoption?

A I mean, it's a general, uh, phenomenon in, uh, in tech that you always, uh, over, overestimate the speed, but underestimate the impact. I think it's probably occurring today. It's slightly different, uh, in the sense that there's some, uh, executive support for, uh, pushing generative AI solutions, even in Europe. So there's some delay compared to the U S market for sure. Uh, but it's not, I wouldn't say it's, uh, it's, it's very significant, uh, It's one year maximum in terms of delay. The challenge here is that it's a technology that is, that can take many forms. And so trying to focus on some specific thing that you can bring to the market that have AI in it, uh, is a prioritization challenge. And so you need to be very strategic around that. And it's, I don't think this is super easy for enterprises. Generally, it will become easier once they try out like a off the shelf solutions a bit more. Uh, once they realize that there are some developer platforms that allows to do it, uh, without hiring, uh, very expensive and hard to find AI scientists in house. Uh, and so we expect that this is. Going to accelerate in the, in the coming years.

AI assessment note: “I think it's probably occurring today. It's slightly different”

Q When you were raising money, was it very different speaking to European investors versus US investors?

A I guess in the seed round, no, it wasn't that different because it was a seed round, uh, for the series A, which was a bigger round. It was, uh, uh, we, I mean, uh, European funds were unstructured to, uh, do the kind of deal that we were proposing. Uh, so we didn't even have a lot of conversation cause they, they just couldn't get their head around the investment that needed to be made, uh, as well as we were a pre-revenue company. Yeah. I think what is lacking and it's related to, to the ecosystem part, uh, in, in, in Europe, our, our growth funds, uh, that are able to, uh, take huge bets, uh, with lots of conviction and that in some should improve over time, especially if, uh, If we manage to, uh, to use a European wealth and channel it more into that growth funds, uh, than it is today.

AI assessment note: “in the seed round, no, it wasn't that different”

Q Can I just ask sufficiently uncoupled, do you not lose efficiency or is there not a leakage between those silos and it creates actually inefficiency by having such silos?

A You have to share some things. So you share the infrastructure, you share the code base, uh, you share findings. Uh, but you know, when you involve, we are doing general purpose models. In general purpose models, you need to evolve them in different directions. So you need to make them speak different languages. You need to make them be able to code, be able to do mathematics, be able to reason. Uh, you need to add multi-modality to them. All of these things are loosely coupled. It's useful if you use the same framework for optimization, for data, for training, but, uh, you don't want to have your team spend their entire day in meetings for coordination. And it's actually pretty hard to figure out. Uh, and, uh, I think so far we've managed to scale it relatively well. Although the team is only 25 people, so that's actually not super challenging. Uh, it will become more and more of a challenge. Uh, but yeah, that's the, that's what I remember from DeepMind. It worked very well at the beginning. Gemini was a bit too slow, and I think they recovered sufficiently well since. Uh, but, um, yeah, that's the, Uh, we, we have optimized the team, uh, to be as fast as possible and to ship as fast as possible.

AI assessment note: “You have to share some things. So you share the infrastructure, you share the code base”

Q So I asked Sam Altman this question. What is the end state for the model landscape? Most people say, ah, it'll become commoditized. And actually there'll be 12 players and it'll be a race to the bottom. What is the end state for models in your mind? And how do you think about the commoditization question?

A I think the end state is, uh, to have a more developed, uh, well, uh, more features on developer platforms that allows to do customization that allows to Make low latency models that serve a certain purpose that allows to evaluate them and to improve them over time. And so the model is only a, like a tiny part. I mean, it's a central part, but it remains a tiny part of an application. And what you want to do across time and when you deploy an application, uh, that you're exposed to users, you want to ensure that it works, ensure that is, that its latency reduces over time, ensure that its quality increases over time. And so I think that's the, the end state is models are effectively going to be A starting point for any AI application developer. Uh, they need to be surrounded by tools, by, um, lifecycle management, uh, platform basically, and that's the one thing that, uh, we started to build. Like general purpose models are a bit undifferentiated, but the differentiation that you need to create for, uh, for your application comes from the data you put into it, the user feedback that you gather and the intelligence that you have to figure out what the application should be doing. And that is not commoditized at all. There's no, Recipe that, uh, allows to go from a A general purpose model to model that is super good and better than all of the others at your specific task. And I t…

AI assessment note: “General purpose models are a bit undifferentiated, but the differentiation... is not commoditized at all.”

Q If you could call yourself up to like the night before you became CEO and founded Mistral and give yourself some advice with the now knowledge that you have, what would you say to yourself, Arthur?

A Maybe stage a bit more the product development and go to market development. Uh, we did start the go to market motion at the time where we had absolutely nothing to sell. Uh, it did work out, uh, did create some brand awareness despite the absence of anything. Um, but, uh, I think it might have been slightly simpler, uh, to state things maybe a little more, uh, developing the product a little before developing the go to market. But since it's such a fast moving field that we did start everything a bit together with some organization that was a bit lacking. And now we are, uh, we are solidifying it, uh, on the flight. Uh, so it's, it has worked out. It hasn't been optimal for sure. Um, and so in hindsight, you can always give you, I could give me like a few tactical advices on who to hire when generally, I think the strategy we had one year ago hasn't changed much. Uh, we did realize that we would need more capital. We did realize that we would need, uh, a strong product. Uh, and that, uh, we could not operate only from Europe, and that we needed to go to the US very quickly. These were findings that we did on, on the way. Uh, I don't think they would have helped, it would have helped that much to know it a year ago.

AI assessment note: “Maybe stage a bit more the product development and go to market development.”

Q Sam and Brad said the other day that models just aren't actually that good any, like yet, and they need to improve a lot in quality. What are the largest constraints or bottlenecks on model quality today and what needs to change for them to improve?

A I think the data quality is, is a constraint. Uh, how do you ensure that the model, how do you leverage the entire world knowledge and ensure that the model follows a certain path toward learning more and more complex things? Uh, that's a very important part. And I think it has been a neglected part. There's obviously compute, but, uh, given the amount of data you have, uh, we have at hand, uh, compute is already running into, uh, is no longer the bottleneck. The bottleneck is more the data at that point. You should look at text to text models. Uh, and so the question is, how do you refine the data, and how do you feed very high quality data to the model itself, uh, in order to improve it over time? And I think in, in that setting, it becomes a bit You, you, one bottleneck that is associated to, uh, bringing better model performance is the question of how do you evaluate these performances? You need to have very good evaluation that targets very specific topics. Like you want the model to be good at, uh, helping diagnosis in, in, in mid, uh, in hospital, but in French. And oftentimes you're a bit out of domain compared to the data you have. And that's where you, you should identify a gap and you should try and fill it out. Uh, so the pushing the model capabilities become, uh, like also a question of mapping where they're failing and figuring out ways of improving it. So for ins…

AI assessment note: “I think the data quality is, is a constraint... compute is no longer the bottleneck”

Q and I'm pleased that you just said that there will be value accrued at the application layer, because I look and I worry that bluntly everything is going to get steamrolled by some of the players that we mentioned. How do you answer the question of will value accrue at the application layer? And for me as an investor today, Arthur, you know me, how would you advise me genuinely?

A There's two opposing directions. So the first is that the models are getting better and better. So it means that Creating a verticalized application as long as you have the data for it and a good understanding of the use case you're facing is going to be easier and easier if you have access to the tools that facilitate it. Uh, so that's the first aspect, uh, which would make me think that, uh, the application layer is going to grow thinner and thinner. But then there's also the fact that the models are, are getting, uh, cheaper and cheaper because we managed to compress them, uh, because we make a lot of improvement on their efficiency, and so that means that effectively this plus the competitive pressure there is on the model layer means that, uh, the price around the model, the dollar per intelligence unit, let's say, is definitely going to reduce. So, uh, there's these two aspects of growing ability, Compressed price, uh, which on one side says that the application layer is going to grow thin, and on the other side says that the model part is going to grow thin. So for us, the, the approach that we are taking is that, is to assume that the model part is still going to be, uh, big enough and that we need to build this platform on top of that, uh, because that's where We are going to enable all of the vertical applications that will be interesting for humanity.

AI assessment note: “There's two opposing directions... on one side says that the application layer is going to grow thin”

Q Who's doing the most margin at the moment?

A NVIDIA is, at that point. The cloud providers are pretty much at cost, uh, LLM providers, we're not at cost, uh, hopefully, but the margin that, Are known to be lower than the typical software margins. AI application makers. Uh, some of them, the one that are most used seems to be doing a pretty good margin. Uh, I think it's going to be quite a moving space. Uh, as I've said, the, the capacity of models makes the, The cost of making an application, uh, lower and lower. I don't think there's any way in which, uh, the marginal cost, uh, and, and the margin of, uh, the most important part of, uh, of that technology, which is really the foundational layer, uh, becomes zero because otherwise there's definitely going to be, uh, I guess, uh, fairness problem.

AI assessment note: “NVIDIA is, at that point. The cloud providers are pretty much at cost”