method · first seen 4 Oct 2026
AI safety via debate
Also: debate, AI debate, debate-based oversight, debate as a training protocol, debate framework for alignment
A scalable oversight approach where two AI agents argue opposing positions to help a judge find the truth.
Mentions per week
PapersCodeDiscussion
Earliest sources
- AI Safety via Debate is Compromised by Cognitive BiasesPapers · 4 Oct 2026 · Gefei Liu, Sonya Rashkovan, Sophia Lloyd George, Isaac Sheidlower, Serena Booth
- Humans Correct Their Judgments Through Debate, but Weaker AI Models Do NotPapers · 20 Oct 2026 · Maria Victoria Carro, Denise Alejandra Mester, Facundo Nieto, Oscar Agustin Stan…
- Efficient Scalable Oversight via RewindingPapers · 29 Oct 2026 · Ziyi Guan, Yael Tauman Kalai, Ron Rothblum, Zoe Xi
- Rational Alignments of Superintelligences via DebatesPapers · 12 Nov 2026 · Thanawat Sornwanee
- Towards a Theory of Learnability via DebatePapers · 12 Nov 2026 · Tosca Lechner, Vinayak Pathak, Sivan Sabato, Ruth Urner
Latest sources
- Towards a Theory of Learnability via DebatePapers · 12 Nov 2026 · Tosca Lechner, Vinayak Pathak, Sivan Sabato, Ruth Urner
- Rational Alignments of Superintelligences via DebatesPapers · 12 Nov 2026 · Thanawat Sornwanee
- Efficient Scalable Oversight via RewindingPapers · 29 Oct 2026 · Ziyi Guan, Yael Tauman Kalai, Ron Rothblum, Zoe Xi
- Humans Correct Their Judgments Through Debate, but Weaker AI Models Do NotPapers · 20 Oct 2026 · Maria Victoria Carro, Denise Alejandra Mester, Facundo Nieto, Oscar Agustin Stan…
- AI Safety via Debate is Compromised by Cognitive BiasesPapers · 4 Oct 2026 · Gefei Liu, Sonya Rashkovan, Sophia Lloyd George, Isaac Sheidlower, Serena Booth