The Bayesian Conspiracy 3 – Interview with Eliezer Yudkowsky
13 snips
Mar 16, 2016 Eliezer Yudkowsky discusses Arbital, his project for making difficult ideas easier to explain, and the tangled challenge of communicating AI alignment. He reflects on self-deception, religious upbringing, writing and the evolution of his fanfiction, then explores prediction, journalism, superintelligent AI, genies, and what could make a utopia genuinely fun.
AI Snips
Chapters
Transcript
Episode notes
Early Religious Ritual Sparked Skepticism
- Yudkowsky described a formative moment at age five or six when being told to pray in Hebrew despite not understanding felt absurd and made him a 'heretic' to the taught ritual.
- That early rejection of authoritative instruction seeded a lifelong habit of challenging authority and testing beliefs.
Use Wiki Pages To Untangle Longform Ideas
- If you blog or teach complex ideas, use a wiki linked to blog posts to avoid a tangled web of cross-references.
- Yudkowsky would create wiki pages for concepts and link blog posts to them instead of chaining posts, improving navigability.
Document AI Alignment Fundamentals Now
- Prioritize writing a comprehensive, accessible brain dump on AI alignment if you can — document orthogonality, decision theory, and instrumental goals.
- Yudkowsky says LessWrong only introduced the problem; a fuller, linked resource would transfer tacit field knowledge.