alt Hacker News>But in this case Google itself is putting out slanderous information it has created itself. So Google in my mind is left holding the buck.
Wouldn't this basically make any sort of AI as a service untennable? Moreover how would this apply to open weights models? If I asked llama whether someone was a pedophile, and it wrongly answered in the affirmative, can that person sue meta? What if it's run through a third party like Cerebras? Are they on the hook? If not, is all that's needed for AI companies to dodge responsibility is to launder their models through a third party?
> Wouldn't this basically make any sort of AI as a service untennable
If the service was good enough that you'd accept liability for its bad side effects,no?
If it isn't good enough? Good riddance. The company will have to employ a human instead. The billionaires coffers will take the hit, I'm sure.
E: > If not, is all that's needed for AI companies to dodge responsibility is to launder their models through a third party?
Honestly, my analogy would be that an LLm is a tool like a printing press. If a newspaper prints libel, you go after the newspaper, not the person that sold them the printing press.
Same here. It would be on the person using the LLM and disseminating its results, rather than the LLM publisher. The person showing the result of the LLM should have some liability if those results are wrong or cause harm