You’ll get an answer with all of the linguistic flavor of Gödel’s texts and absolutely none of his insights or reasoning.

LLM is a language model, not Deep Thought from the hitchhikers guide.

It works on embeddings, which are concepts. So it would at least be manipulating a similarly weighted network of concepts, though much shallower. It would be an answer from his literal domain of discourse. It would be evocative rather than authoritative.

What you describe would be literally worse than nothing. The use of those concepts would make about as much sense as the use of topological concepts in a Lacanian text. A typewriting monkey in a trenchcoat.

I can understand the downvotes, but at the same time it does makes sense, especially given the target is a logician.

> For a possibly not useless proxy, train an LLM on his oeuvre and ask it.

all major LLMs are likely already trained on his works, so you can already ask: "Hi, ChatGPT, pretend to be Kurt Godel"