According to a post on LessWrong, AI agents often underperform in optimization tasks not due to fundamental capability gaps, but because they're being under-elicited—insufficiently prompted and instructed. Researcher zef presents a case study suggesting that better techniques for extracting agent potential through improved prompting could yield substantially higher performance, hinting at a gap between theoretical and practical capabilities.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton