Post #2361990

RE: https://fediverse.zachleat.com/@zachleat/116534277384407110 I'd be wary of *anything* that can be *convincingly wrong* regardless of age bracket. Since an LLM is not a person, it cannot be "confident". If, however, the output convinces a person to accept the LLM output as true, it's the *person* who makes the mistake. And the blame must be split between the people who lied to convince the user that the LLM could be reliable (even while hedging the UI with warnings) and the gullible user who accepted their lies. #LLM #AI