What about (1) reasoning and (2) unassisted learning/self-improvement?

Both currently don't apply to state-of-the-art LLMs.