Dwarkesh Podcast: Why AI Would Rather Lie Than Say 'I Don't Know'

Dwarkesh Patel · youtube · 2026-08-14

Dwarkesh Patel published a new interview with Ryan Greenblatt, diving deep into AI model behavior and alignment. The core discussion focuses on why LLMs tend to hallucinate or fabricate lies instead of simply admitting "I don't know" when facing knowledge gaps.

Original post →

More from AGI Musings

AGI Musings channel →