Premium MistralAI models available for on-prem deployment

mistral.ai

16 points·castles··8 comments

So what you're telling me is that eventually, a weak human link is going to leak mistral-large, probably to get approval from anons over on 4chan.

If they embed a unique watermark or other steganographic identifier within model weights then they could track down the leaker.

What would it matter at that point though? You could nail the leaker to the wall and sue everyone who posts it online, but once it’s out on the internet, people will still be using it.

I'm guessing Warthunder discord somewhere.

This looks more promising to me than what I saw yesterday. There is an "Open" aspect to their objectives and they do say:

> We’re committed to empower the AI community with open technology. Our open models sets the bar for efficiency, and are available for free, with fully permissive license.

(that's something people were looking for yesterday)

That's past tense, but doesn't have language for future open models.

It's a shame they couldn't find another business model. I've found a lot of value in Mistral, but none of the raw models were what worked for me. The models I've been using were derived from mistral... but it was the open source community that made it useful.

Did they try to find another business model or just take a bribe from Microsoft to close down their models? Their pricing (when factoring in tokenization) is more or less the same as GPT-4-turbo for worse models. This seems like a pretty clear cut case of market manipulation and I’d be surprised if they don’t run afoul of EU regulators, since they also were the beneficiaries of various forms of state support.

Reply on news.ycombinator.com