"Emergent capability" claim mocked; Fleet hiring researchers to automate mech interp with frontier models

sebkrier · x · 2026-11-03

Two threads: AlexKrentsel's claim that weight-space programming is a step-change "emergent" capability in the newest models drew mockery for falling into the "emergence" trap. Separately, Fleet researcher Michael Sklar (ex-Anthropic interp resident) is recruiting to "mechanize mech interp" — frontier-model-aided interpretability research whose failures become RL training tasks; Fleet sells RL gyms to every frontier lab.

Original post →

More from Fun

Fun channel →