New paper: output similarity doesn't imply process similarity in LLMs
ValerioCapraro · x · 2026-10-13
The author argues one of the biggest mistakes in the AI debate is conflating output similarity with process similarity: a machine behaving like us doesn't mean it thinks like us.
A new paper illustrates the point: humans and two LLMs repeatedly chose between standard and slower delivery, and all three responded similarly on the surface. But deeper analysis shows the underlying decision processes differ, demonstrating that similar outputs do not imply similar underlying mechanisms. Paper link included.
Related event: Zhejiang Study: LLMs Decide Like Humans but Adapt Differently(2 posts)→
More from AGI Musings
- Satya Nadella spotlights essay: treat frontier models as insider risks in the superintelligence era — luisdans · 2026-10-13
- Nate Silver and Thompson clash: labs say they're 'not ready for RSI' while trying to build it — wfithian · 2026-10-13
- Ask Claude for its preference instead of a recommendation and it turns non-moralistic — austinc3301 · 2026-10-13
- Replies are turning into AI slop: screenshot shows bot-style engagement comments — martyamark · 2026-10-13
- Dimon warns on AI agents tapping bank accounts; commentator says he's preventing a bank run — robleclerc · 2026-10-13
- AI-age propaganda may not need shared beliefs — just different emotions, same outcome — CyborgWriter · 2026-10-13