Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Mistral's Shieldstral: 3B open-weights model for multimodal moderation

mistral.ai

Post reply on HN
Print view View original

31–40 of 154 posts

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#31

Post by baq »

LunaSea wrote: post #26
baq wrote: post #22

Earlier quoted context omitted.

Distilling is unsafe from export control perspective - Chinese models are poisoned by US frontier distillation and a case can be made that the US won’t like distilling what they may consider transitively theirs, which they will the moment you’re anywhere near competitive.

US judges have already rules that output of an LLM can't be copyrighted so not sure what would prevent Chinese companies to use said output for distillation purposes.

Note I didn't mention copyright

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#32

Post by cinntaile »

petcat wrote: post #10
mosura wrote: post #8

Someone should use this to do the exact opposite of the intention: filter for “offensive” content, and boost it or collate it into a newsletter/email blast for people of culture. You have to give it to Mistral they do at least know what the market near them says they want right now. The great problem is in a few years of this that market won’t be worth anything. Edit to add, you could also add this to an AI workflow…

> they do at least know what the market near them says they want right now It does seem to be a very European approach to AI that their flagship AI lab is just making models that do nothing other than monitor and moderate internet content. I guess they know that the EU AI Act, Chat Control, etc are going to cause a lot of companies to need this kind of compliance.

Another commenter already mentioned that it's more likely a lack of compute and funding that forces their hand to focus on niche tasks.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#33

Post by mcintyre1994 »

mosura wrote: post #8

Someone should use this to do the exact opposite of the intention: filter for “offensive” content, and boost it or collate it into a newsletter/email blast for people of culture. You have to give it to Mistral they do at least know what the market near them says they want right now. The great problem is in a few years of this that market won’t be worth anything. Edit to add, you could also add this to an AI workflow…

> filter for “offensive” content, and boost it or collate it into a newsletter/email blast for people of culture.

I think that's the main service that xAI provide for X.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#34

Post by pwython »

I've had dreams of building something in the image sharing or social platform realm, but stopped short of planning because of obvious content moderation responsibilities. This seems to be a realistic, cost effective solution to that one piece of the puzzle.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#35

Post by Me1000 »

NitpickLawyer wrote: post #30
LunaSea wrote: post #26

Earlier quoted context omitted.

US judges have already rules that output of an LLM can't be copyrighted so not sure what would prevent Chinese companies to use said output for distillation purposes.

> already rules that output of an LLM can't be copyrighted Mind sharing such cases? I'm not aware of any so far. There's the one with images, but that's commonly miss-understood, that case was ruled on a technicality (i.e. copyright needs to be attributed to a person, not a model)

https://www.copyright.gov/newsnet/2025/1060.html

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#36

Post by kergonath »

pwython wrote: post #34

I've had dreams of building something in the image sharing or social platform realm, but stopped short of planning because of obvious content moderation responsibilities. This seems to be a realistic, cost effective solution to that one piece of the puzzle.

I am not sure how reliable it is in the real world. Also, in terms of liability, I don’t know how effective it would be to satisfy various regulations compared to a human moderator team.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#37

Post by kergonath »

mosura wrote: post #8

Someone should use this to do the exact opposite of the intention: filter for “offensive” content, and boost it or collate it into a newsletter/email blast for people of culture. You have to give it to Mistral they do at least know what the market near them says they want right now. The great problem is in a few years of this that market won’t be worth anything. Edit to add, you could also add this to an AI workflow…

> The great problem is in a few years of this that market won’t be worth anything.

To be fair, we don’t know how much resources they put into this and how much of a distraction it was. If it was quick enough to train or fine tune and it brings them valuable experience for the next models, it could well be worth it in the long run even if there is no direct successor.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#38

Post by pebbly_bread »

kergonath wrote: post #37
mosura wrote: post #8

Someone should use this to do the exact opposite of the intention: filter for “offensive” content, and boost it or collate it into a newsletter/email blast for people of culture. You have to give it to Mistral they do at least know what the market near them says they want right now. The great problem is in a few years of this that market won’t be worth anything. Edit to add, you could also add this to an AI workflow…

> The great problem is in a few years of this that market won’t be worth anything. To be fair, we don’t know how much resources they put into this and how much of a distraction it was. If it was quick enough to train or fine tune and it brings them valuable experience for the next models, it could well be worth it in the long run even if there is no direct successor.

Also, it sounds like the kind of thing that sells. Any company with a customer support chat is a potential user of this model, any large company may be interested in getting a mistral installed set-up for handling without needing to send client info over the web. Installing those kinds of local systems seems to be what butters Mistral's bread at the moment.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#39

Post by kergonath »

petcat wrote: post #10
mosura wrote: post #8

Someone should use this to do the exact opposite of the intention: filter for “offensive” content, and boost it or collate it into a newsletter/email blast for people of culture. You have to give it to Mistral they do at least know what the market near them says they want right now. The great problem is in a few years of this that market won’t be worth anything. Edit to add, you could also add this to an AI workflow…

> they do at least know what the market near them says they want right now It does seem to be a very European approach to AI that their flagship AI lab is just making models that do nothing other than monitor and moderate internet content. I guess they know that the EU AI Act, Chat Control, etc are going to cause a lot of companies to need this kind of compliance.

> It does seem to be a very European approach to AI that their flagship AI lab is just making models that do nothing other than monitor and moderate internet content.

Well, first Mistral is French more than European. This might be a difficult distinction to make from the US but their approach is quite different from e.g. typical German companies.

Then, this is just a small model they release on the side. If that’s your benchmark, they released somewhat recently Voxtral, Voxtral transcribe, their OCR model, and Leanstral. I don’t think you can get much insight on their culture from this kind of release.

Re: Mistral's Shieldstral: 3B open-weights model for multimodal moderation

#40

Post by kergonath »

baq wrote: post #22
yborg wrote: post #16

Earlier quoted context omitted.

By this logic the Chinese should have just given up and let the American AI companies have the market because they were so far behind. I'm sure Europe has the capability to distill other people's frontier models to catch up if they wish to do so.

Distilling is unsafe from export control perspective - Chinese models are poisoned by US frontier distillation and a case can be made that the US won’t like distilling what they may consider transitively theirs, which they will the moment you’re anywhere near competitive.

> Distilling is unsafe from export control perspective

That is not the direction American judges are taking. Right now, they are saying that LLM output cannot be copyrighted. And if looting copyrighted works for training is fair game, I really don’t see how one could argue that learning from other LLMs is not.

Post reply on HN