There are problems LLMs can't solve (a proof) If there is a problem that is out of a large language model's training distribution and requires greater context than the maximum window, that problem would be fundamentally unsolvable by LLMs.
