Position: LLMs Can't Jump

	Position: LLMs Can't Jump(openreview.net)
	290 points by theanonymousone 1 day ago \| 192 comments
	tl;dr: Summary not available
	HN Discussion: ↑Language is inherently lossy, supporting limits on LLM understanding and creative leaps ↓The paper's argument is unrigorous opinion lacking quantitative evidence or testable claims ↓Recent LLM achievements (e.g., OpenAI unit distance result) already contradict the paper's claims ↓The limitation is the harness/environment, not LLMs themselves; embodiment enables jumps ↑LLMs fundamentally can't 'not know' or truly jump because they're search-like by nature