alt Hacker NewsThis is a truly important paper. It formalizes the intuition that many in the field have. We can stop wasting time doing formal analysis of LLMs. If you have a problem that requires formal verification, don't use an LLM. You can use an LLM to help you build such a system, but the LLM can't be the system.
Replies
suddenlybananas • yesterday at 10:22 PM
I don't really see the relationship to your comment and the paper's content. Could you elaborate a little?
➕ show 1 reply
I'm not sure that's a great takeaway. Lots of problems are undecidable and have reasonably effective solutions in practice (e.g. finding bugs -> testing, static analysis, etc). Mind you, I don't expect we'll find anything like that for transformers, but there's a surprisingly large gap between what's possible in general and what's possible in the cases we care about.