Let's talk about Recursive Self-Improvement

​AI is now helping build the next version of itself. METR just spent 11 weeks inside Anthropic and found AI progress there is running roughly twice as fast as a year ago, mostly because AI is doing more of the work. Let’s dig into what that means.

​The plan

• ​Level-set (all together): What RSI is, why it matters, and the latest updates

• ​Breakouts:

  • ​Deep-context group: Self-organized technical sessions on AI control, interpretability, or whatever area you choose

  • ​Newer-to-this group: Read a couple of short pieces together, discuss, and learn from each other

• ​Regroup: Share what each group discussed

​📖 Required reading (~15 min)

• ​Is Anthropic on the cusp of RSI? Inside the redacted METR report, 80,000 Hours

​Optional resources

• ​AI is dramatically accelerating AI research right now, Ajeya Cotra

• ​Buck Shlegeris’s take

• ​METR’s full report

• ​How AI-driven feedback loops could make things very crazy, very fast, Benjamin Todd (great for newcomers)

​Thank you Mox SF for hosting us! Mox is a coworking & community space for do-gooders in AI Safety. Check out our memberships here.

​Brought to you by AI Safety SF: a community of people in San Francisco who care about making sure AI goes well for humanity.