Post #1472689
2026-04-14 07:21 UTC
@markgritter Is (b) true? I'd think it would do fine with the grammar of the new language and a good codebase in any other language. I'd imagine this could even be done just within the context window without even re-training the LLM. But I don't actually know.
Replies (1)
-
@markgritter@mathstodon.xyz 2026-04-14 07:58
@poleguy The paper I linked downthread tries several approaches! It does not look like (in the current models) a description of the language is good enough. An interactive session does help somewhat in their experiment. I would like to see a non-Esolang example, although the effect seems very strong there. While (b) is certainly a questionable assumption, I think it is currently borne out by this and similar results -- the models get good by having training sets that are relevant, rather than developing transferrable knowledge.