abstract inductive reasoning requires dynamic learning in unfamiliar contexts. All the benchmarks which measure this specifically (in particular, task compositi

abstract inductive reasoning requires dynamic learning in unfamiliar contexts. All the benchmarks which measure this specifically (in particular, task composition) see LLMs fail catastrophically. There is quite a lot of research published on these limitations now.