I don't think there are problems unsolvable in principle. Given a good enough specialist and some amount of time, it's possible to guide an LLM to the solution eventually.

The problem is that people often can't recognize whether they are getting closer to the solution or not, so iteration breaks.

about unsolvability, it depends. In some contexts there are undecidable problems, like the halting problem, but I get your point.

yes that's my point. The article seems to suggest that the opposite is true, that without a specialist you can solve special problems with an unlearned user and a LLM.