Comment by dwedge
8 hours ago
I've tried Mistral a few times, at first it seemed promising (though lagging) but at some point it seems like they stopped focusing on AI and shifted their focus to being a mouthpiece for EU policy and pushing for regulation. I can't really take any of their announcements seriously anymore.
A couple of weeks ago they were calling for a European AI tax to pay creatives.
Sounds like the tax on recordable CDs.
https://en.wikipedia.org/wiki/Private_copying_levy
Ridiculous, and it didn't take money from the right people or give it to the right people. I would expect the same from an AI tax.
Recordable CDs involved individuals making copies. AI is run by a couple of dozen people who give full access to other people's work, metered by the syllable.
It was never legal for massive corporations to record other people's work on CDs and sell them; that's the opposite of copyright. The comparison is absurd.
I lived and studied in France. So it's only natural for me to try out Mistral's every major updates. I had the same sentiment with you, that their models were just a bit lagging. But on the other hand, I understand their shift. Their value is not in SOTA coding, math, or puzzle solving performance in my opinion. They will catch up. To me it makes sense that they focus on something else, how to scale through talents and propagate their models with policies in Europe.
They are making some very good specialised models, like Voxtral
Why do you say they "stopped focusing on AI"? I see a pretty consistent release of pretty good products - particularly in speech and OCR.
I used to use Mistral OCR, but found it was better just to write a program that sent the documents to Claude Sonnet to OCR instead. Claude is far better quality, better formatting and fewer errors.
I'm also using Voxtral TTS to try to replace OpenAI. It "works", but I've had problems with volume levels being radically different between different audio chunks. It doesn't seem to "understand the full text" the way OpenAI's voice models do, which can be more expressive. Voxtral sometimes sounds robotic in the reading. And some Voxtral TTS output contains music in the background occasionally, which suggests their training corpus isn't that clean. Try generating a personalized news podcast, and the intro may occasionally sound like the music for BBC News underneath....
As for not focusing on AI, there's this interview in the Big Technology Podcast 2 months ago, where the Mistral CEO says their main focus is on helping companies fine-train models for internal use, over being a general model builder.
https://www.youtube.com/watch?v=xxUTdyEDpbU&t=1357s
"I sent money to the god knows how many trillion parameters fully closed source machine built on billions of dollars and it worked better than the model that I can self host from the guys next door"
yeah, no shit ? All you're saying is that you're happily locking yourself in to models you have zero control over and that Anthropic can fuck you over at any time.
However, yes, Mistral is not in the business of providing you with a perfect, general purpose model. They fine tune from their base models for specific tasks.
1 reply →
I used their OCR against a few hundred page PDF that was printed text but missing the OCR. It cost me $5 and was useless, it did worse than tesseract. That's how all my experience with mistral is
“European AI tax to pay creatives”
Love that idea.
I mean we already pay a tax when buying phones or storage because it's assumed we'll use it for piracy so why not.
If they want to improve but find that European regulations are the main obstacle, it makes sense they focus on that.
Maybe I wasn't very clear. It seems like they are in support of those regulations. They seem like an EU mouthpiece