The Inside View
Transcript: https://www.alignmentforum.org/posts/NFfZsWrzALPdw54NL/the-inside-view-3-evan-hubinger-homogeneity-in-takeoff
Outline:
00:00 Evan’s background @ MIRI & OpenAI
3:29 Coconut & functional programming
6:56 Homogeneity in AI takeoff
14:08 Reproducing SoTA & openness in multipolar scenarios
29:16 Quantilizers & operationalizing strategy stealing
42:16 Risks from learned optimization & evolution
52:29 Learned optimization in Machine Learning
1:04:51 Clarifying Inner AI Alignment terminology
1:13:33 Transparency & Interpretability
1:25:26 11 proposals for safe advanced AI
1:39:13 Underappreciated problems in AI Alignment & surprising advances in AI