Hacker News Reader

Comment by vintagedave | Hacker News Reader

> Mistral is just not competitive enough

Does anyone know why? I was really excited when they emerged, but their models and targets don't seem to be quite in the same market.

View on HN

Their target market is completely different. Anthropic and OpenAI try to build general AI that wins on all the benchmarks by throwing ungodly amounts of money at it.

Mistral focuses on long term b2b contracts and their proposition is that they fine tune their model to your needs with an added bonus of 'not dependent on America' in a politically tumultuous time.

An so like if a business wanted to home in on one very specific use case that could be hyper optimized by SFT, had really good support for updating and adding new features, on-Prem etc. that’s the kind of market they are in?

Another added bonus is that they offer a clear "self hosted" if you want it, you can get the exact same product without sending your queries to them, it's costlier and you need the hardware sure, but between the economic espionnage aspect, the sovereignty aspect, the data safety aspect ... This has teeth in europe.

Lack of capital and (probably) lack of willingness to mass distill Anthropic and OpenAI.

What would happen if they mass distilled one of the really large local models like GLM 700b or deepseek 1.6t?

At that point you might as well just host them yourself.

That's not how the innovation works

Innovation is teaching your model on stolen data from literally everywhere but other models.

Innovation is a pretty neutral concept. It doesn’t care about things like “what if my model learns from other models” as opposed to “what if my model learns from data I painfully curated” if the model progresses the same.

Those already exist.

Most probably lack of capital and talent. At the end of the day they have to compete with other giants for the chips to train the models.

I wouldn't be surprised if they had new models up their sleeve. Could be wrong of course.

I’m pretty sure they have new models, but not better ones

capital and talent is the same in this context

there's no shortage of talent in Europe or France, it's just an issue of available capital

What I meant was top talent. US is still the top destination for top AI talent in general

A large contingent of the top AI people is French, the Mistral founders worked at Meta and Google before coming back to France. The real issue is capital, French salaries are shit

"there's no shortage of talent in Europe or France, it's just an issue of available capital"

This is more complicated than you paint it. Countries like UAE have enough capital to throw at things and little-to-no taxation, yet they don't attract as much talent as they would like to.

Preexisting centers of excellence like Silicon Valley are attractive for young talented people precisely because a lot of older talented people are already there. The same reason why a young talented painter in 15th century would prefer Florence to some rich, but boring place elsewhere.

You can only really do a meaningful work in a "heavy" field by tightly cooperating with others, and physical proximity still matters.

the older talented people are there because of money

there's nothing special about the US, it's money

Reply on news.ycombinator.com