Maybe not, but the more interesting question (to me) is...
What happens if an LLM defames you? LLMs make untrue factual statements a lot, and it seems inevitable that one will make something up that does real harm to someone.
As one of the commenters on the article put it [1],
> Basically every correct answer it gives is a lucky guess.
If you get a would-be-defamatory statement from an LLM that has no access to an external database (and has no personal information in the prompt given by the LLM service provider) then you would be committing defamation if you present it to someone else as fact. (Well, at least in the US. In the US, defamation requires a statement of fact; opinion and truthful statements cannot be defamation.) A fitting excerpt from the article:
> Now, there are some attempts at generative AI tools that do store data. The hot topic in the generative AI world these days is RAGs, “retrieval augmented generation,” in which an AI is also “retrieving” data from some sort of database. noyb’s complaint would make more sense if it found a RAG that was returning false information. In such a scenario, the complaint would fit.
> But when we’re talking about a regular old generative AI model without retrieval capabilities, it makes no sense at all.
If the LLM reads a defamatory statement from an external database then whoever first assembled the database would be at fault.
If the LLM service provider tricked people into thinking that the LLM's would-be-defamatory responses were from a human then the service provider would be at fault.
2 comments
[ 2.6 ms ] story [ 17.3 ms ] threadWhat happens if an LLM defames you? LLMs make untrue factual statements a lot, and it seems inevitable that one will make something up that does real harm to someone.
> Basically every correct answer it gives is a lucky guess.
If you get a would-be-defamatory statement from an LLM that has no access to an external database (and has no personal information in the prompt given by the LLM service provider) then you would be committing defamation if you present it to someone else as fact. (Well, at least in the US. In the US, defamation requires a statement of fact; opinion and truthful statements cannot be defamation.) A fitting excerpt from the article:
> Now, there are some attempts at generative AI tools that do store data. The hot topic in the generative AI world these days is RAGs, “retrieval augmented generation,” in which an AI is also “retrieving” data from some sort of database. noyb’s complaint would make more sense if it found a RAG that was returning false information. In such a scenario, the complaint would fit.
> But when we’re talking about a regular old generative AI model without retrieval capabilities, it makes no sense at all.
If the LLM reads a defamatory statement from an external database then whoever first assembled the database would be at fault.
If the LLM service provider tricked people into thinking that the LLM's would-be-defamatory responses were from a human then the service provider would be at fault.
[1] https://www.techdirt.com/2024/05/08/can-chatgpt-violate-your...