Mistral: The 9-Month-Old AI Startup Challenging Silicon Valley's Giants | Hacker News Reader

Mistral: The 9-Month-Old AI Startup Challenging Silicon Valley's Giants

wsj.com

About leaving Google:

> “I think I left just before it got too bureaucratic for me,” Mensch said. “I didn’t want to build opaque technology from within big tech.”

Interesting point of view in light of the ongoing train wreck of bard/gemini/gemma/whatever it's called today thing.

From zero to 2 billion valuation in 9 months is impressive.

It might be cynicism, but "I educated myself and others on Google's budget, and left at the perfect time to grab some AI investment cash with the expertise I gained there" feels more likely. Perfectly reasonable move, but why not just say it?

Even if it's true, it would be a terrible move marketing-wise.

It's good for memoires, around a decade after the hype died down.

> "I educated myself and others on Google's budget"

Do you assume he did not produce anything valuable at Google while working there ? (That's entirely possible, but not certain.)

As the old joke goes:

    A: "What if I train them, and they leave ?"
    B: "Sure. What if you don't train them, and they stay ?"

He's the main lead in developing a model called chinchilla, which at that time provided results as good as those of gpt-3

> Do you assume he did not produce anything valuable at Google while working there

Oh - I'm not saying that. I assume he did, but was also paid to do that.

I'm more commenting on the marketing, rather than the person.

> “I educated myself and others on Google's budget, and left at the perfect time to grab some AI investment cash with the expertise I gained there”

That’s been the Silicon Valley way ever since 1957 when Fairchild Semiconductor was started by the “traitorous eight” who later jumped ship again to found companies like Intel and AMD.

That is normal in companies. I am more worried by the cases of 'I educated myself by using public funds for my research at uni, then went of to create a company and privatize the profits'

Having spent far too long in the annals of higher education, those who would be comparable stewards of growth are few and far between.

I doubt someone who doesn't know that it's profits that get taxed would be able to make much of a company. I wouldn't worry.

"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it." Bill Gates

is this "train wreck" the bad image generation regarding people and races or is there anything more?

That does not affect my usage and Gemini works fine for me.

Imagine going thirty years back and telling people that a leading complaint of our age is:

“This free artificial intelligence doesn’t generate sufficiently accurate pictures of non-existing Nazis.”

Talk about first-world problems. (Back then most people still had the naïve idea that they’d probably rather see less pictures of Nazis, much less spend time looking at ones that didn’t even exist.)

Talk about good timing. The guy was not even doing anything related to LLMs before 2020 and his first relevant publication is in 2021.

> From zero to 2 billion valuation in 9 months is impressive.

Assuming it's actually worth $2b, of course.

I'd argue the fact startups are able to compete in just 9 months suggests there is little to no moat here and others will be able to do the same. I'd also argue there's a lot of speculation around the value of these products right now so it's debatable wherever it's actually worth $2b even if there is a moat. Not saying it isn't, but I don't think anyone really has a good sense for the long-term value of these new AI products / companies just yet. The valuation is mostly a result of speculation and AI FOMO.

But regardless of valuation, it's still insane that Google has become so dysfunctional that they're now struggling to compete with tech offerings from 9-month old startups. And this article does a disservice saying "big tech", because this is specifically a Google problem. Meta has had no problems pivoting their focus towards AI, creating a strategy and releasing interesting products and models.

And it's not even just that Google's products are bad, their announcements always feel rushed and clumsy, and their AI strategy seems to be nonexistent. None of this is a result of a lack of talent at Google, it's all down to crappy leadership and bureaucracy.

Agree. If a competitive model can be trained in a few months, its not clear what is really unique here

They don't want these chat models to hurt their search business. But due to the hype and fomo they are rushing things up. Their models before bard(before the rush) were good. They just need to get back in a good track and good leadership.

Honestly, I don't AI chat is anywhere near competing with search at this moment in time. It will come, but there's no significant near-term risk that AI chat is going to be the place to go for common Google searches like: "what's the weather like tomorrow in London", "Facebook", "Sofas", "funny cat videos".

AI chat is really only useful for a niche area of the search market orientated around non-real-time question and answer type questions. Eg, "who was the greatest US president?"

The valuable search ad space is mostly around transactional searches like "sofas" where a company could place an ad and convert a user looking for a new sofa. No one is using chatbots for this. That's not to say it's not a risk, but I think people are wrong to think that web search and AI chat are directly competing products. AI chat competes with search in a similar way that Alexa and Siri already does – they're useful within a certain niche.

Signs of another tech bubble if you ask me…

I am curious about how they gained access to a huge cluster to train a 7B model from scratch. It seems like they used the EuroHPC cluster, Leonardo, but I am not sure if it was provided by EuroHPC or if they paid for it. If it was a giveaway, what privileges did they have to access such a large cluster?

European tech companies will always be behind the US. They are giving €80,000 to developers in firms valued at $2 billion :) Look at the salary scales of OpenAI and other companies. Yo will see the difference.

You're making two key assumptions:

* How far ahead/behind a company is is entirely a function of developer talent.

* How talented your developers are is entirely a function of their salary.

I strongly believe that neither one of these things is true. We'll see who's right in the coming years.

I found some discrepancies when reading deeper into Mistral that felt a bit peculiar.

If you look at the Wikipedia page for Mistral AI [1] it lists the company as having 3 founders, one from DeepMind & 2 from Meta. This is consistent with what I thought- that it's an AI lab founded by engineers.

But if you go to Cedric O's Wikipedia page [2], it has Cedric O, a politician and former Secretary of State for the Digital economy, listed as a co-founder. The source cited is [3]. Furthermore, if you look at this post about Mistral's lobbying for the AI Act [4], Cedric O is said to be just an investor.

These discrepancies seemed a bit weird and suspicious to me; it felt like they might have been trying to downplay how close their ties were with the French government?

[1]: https://en.wikipedia.org/wiki/Mistral_AI [2]: https://en.wikipedia.org/wiki/Cédric_O [3]: https://sifted.eu/articles/brunch-with-cedric-o [4]: https://www.euractiv.com/section/artificial-intelligence/new...

For a company of Mistral's age, the term founder is quite fluid - people may not have been at the very start (Musk joined Tesla 7 months after it was incorporated)

Others say there were 6 founders, for example:

> Three of Mistral’s six founders, and its technical brains—Mr Mensch, Timothée Lacroix and Guillaume Lample ....

https://www.economist.com/business/2024/02/26/meet-the-frenc... https://archive.ph/2GPNo

So all he has to do is sue Mistral for the founder title.

I don't know anything about the subject but it's worth noting that the article Wikipedia cites for listing the three founders, also specifically mentions Cedric as one of three additional "non-operational" co-founders.

My interpretation from the article cited on Cedric's Wiki page and published 6-7 months after launch is that he joined after and hence isn't one of the three original founders, maybe:

> O explains he was drawn to Mistral a few months ago first and foremost because of the ambition of its founders Arthur Mensch, Timothée Lacroix and Guillaume Lample, Meta and Deepmind alumni.

Not saying you're wrong. But "he's not listed on Wikipedia" is a bit weak? Consider just adding him and the others if you figure out who joined when.

I think you're right and I clearly read too much into the Wikipedia pages, instead of looking for more primary source documents like those others have posted. I will be more careful in the future. Thank you for pointing that out!

Their strategic memo created in April/May last year present a founding team of 6 people actually: https://philippeoger.com/documents/1/mistral.ai-strategic-me...

The memo in itself is a very good read.

Thank you for sharing- upon reflection I should have been looking for primary source evidence like this instead of reading too deeply into Wikipedia pages, as someone else pointed out. I will definitely be more careful in the future.

Could I ask how you would normally go about trying to find memos like this for other companies?

Well, you made a worthwhile comment that triggered an interesting discussion, which luckily I had some knowledge about. I simply brought up new information to you. I don't perceive anything wrong.

I work in data and I'm french, so Mistral.ai has been on my radar a little more maybe. I was familiar with this memo before Mistral even released their first model. It was discussed a lot in France as this was the base for their first seed round that gave them 100 million euros in investment.

More here if you're interested: https://philippeoger.com/pages/ai-scene-in-europe-last-week/

It's interesting that France's intelligence service in particular is known for skill in industrial espionage.

Reply on news.ycombinator.com