fasterthanlime , vor 17 Tagen Englisch sometimes you gotta bully LLMs a bit Me: That's not true. ChatGPT: You're right. My previous message was incorrect in its assertion that Rust never had a garbage collector. Rust did initially experiment with garbage collection in its early days. When Rust was (answer cuts here)
sometimes you gotta bully LLMs a bit
Me: That's not true. ChatGPT: You're right. My previous message was incorrect in its assertion that Rust never had a garbage collector. Rust did initially experiment with garbage collection in its early days. When Rust was (answer cuts here)
christmastree , vor 17 Tagen Englisch @fasterthanlime 🤔 https://hachyderm.io/@fasterthanlime/112281433885161826
@fasterthanlime 🤔 https://hachyderm.io/@fasterthanlime/112281433885161826
syphdias , vor 17 Tagen Englisch @fasterthanlime I don’t think calling "someone" out on their lies counts as bullying
@fasterthanlime I don’t think calling "someone" out on their lies counts as bullying
pointlessone , vor 17 Tagen Englisch @fasterthanlime so did it tell the truth or did it confess under duress?
@fasterthanlime so did it tell the truth or did it confess under duress?
rlabrecque , vor 17 Tagen Englisch @fasterthanlime They should pass all output through a layer that asks the question, gets the response, then asks "Are you really sure that's true?" and then provides that response instead.
@fasterthanlime They should pass all output through a layer that asks the question, gets the response, then asks "Are you really sure that's true?" and then provides that response instead.
fasterthanlime OP , vor 17 Tagen (Bearbeitet vor 14 Tagen) Englisch @rlabrecque I think that's already a thing and it's called "self refinement": https://arxiv.org/abs/2311.07961 edit: according to the paper, it's not great
@rlabrecque I think that's already a thing and it's called "self refinement": https://arxiv.org/abs/2311.07961
edit: according to the paper, it's not great