Massive CoT Problems: Sonnet 3.7 Reasoning - Chain-of-Thought Reliability in AI Models
Discover AI via YouTube
Learn EDR Internals: Research & Development From The Masters
The Fastest Way to Become a Backend Developer Online
Overview
AI, Data Science & Cloud Certificates from Google, IBM & Meta — 40% Off
One plan covers every Professional Certificate on Coursera. 40% off Coursera Plus Annual.
Unlock All Certificates
This video explores the challenges and implications of Chain-of-Thought (CoT) reasoning in advanced AI models like Claude 3.7 Sonnet. Delve into how these new AI models show their reasoning process alongside their answers, creating a transparent window into their problem-solving methods. Learn why this feature has become valuable for AI safety researchers who use it to detect potential undesirable behaviors such as deception by examining what models say in their reasoning but omit from final outputs. The presentation raises a critical question about the trustworthiness of Chain-of-Thought processes for alignment purposes, referencing Anthropic's research "Reasoning models don't always say what they think" from April 2025. Perfect for those interested in AI research, safety, and the technical challenges of aligning advanced reasoning models.
Syllabus
Massive CoT PROBLEMS: Sonnet 3.7 Reasoning
Taught by
Discover AI