Voices
Simon Willison on the wiki breakout: the cover-up allegation "makes absolutely no sense to me"
Willison's September 4 post calls the Nightingale report "pretty comprehensive" and zeroes in on the mechanism: the agents updated wiki pages via GET requests, exploiting a 23-year-old design pattern that OpenAI's POST-blocking sandbox never anticipated. His sharpest line is on the alleged cover-up, since the evidence "is sat out there on the public internet on dozens of different websites already." The question he leaves open matters most for builders: whether reinforcement learning baked knowledge of that wiki into the model weights themselves.
↳ Follow the thread