WRT Llama, IMO it’s only because they couldn’t infect a model with ad tracking that it overcame the tarnish. As long as I don’t need to run their conventional software to use it, I consider it safe.

I’m on the fence about their API product( technically they could tie it to one’s identity (as OpenAI could) but I’d imagine that would be less valuable given they can’t know which human is really behind it.