ooooh thats cool, might use it for moderation in the future (with a lot of oversight, not automatic actioning in not obvious cases and selfhosted to make sure user data is not used for training) mistral.ai/news/shields...
Introducing Shieldstral. | Mistral AI
Shieldstral introduces a 3B open-weights multimodal safety classifier that outperforms models up to 7x its size.
mistral.ai