
LessWrong (Curated & Popular) Critical review of Christiano’s disagreements with Yudkowsky
Dec 28, 2023
Paul Christiano and Eliezer Yudkowsky discuss disagreements on pivotal acts, take-off speeds, and recursive self-improvement in AI. They also explore addressing risks in transformative AI systems through factored cognition, evaluation challenges, imitation learning, unknown unknowns of deep learning, and disagreements on AI development.
Chapters
Transcript
Episode notes
1 2 3 4 5 6
Introduction
00:00 • 5min
Disagreements on Pivotal Acts and Recursive Self-Improvement in AI
04:43 • 5min
Addressing Risks in Transformative AI Systems through Factored Cognition
10:01 • 2min
Evaluation Challenges and Imitation Learning in Alignment Proposals
12:01 • 3min
Unknown Unknowns of Deep Learning and the Safety of Alignment Protocols
14:41 • 2min
Disagreements on AI Development
16:17 • 12min

