Neuralese panic over OpenAI claim sparks pushback: 'just tell independent auditors the architecture'

zetalyrae · x · 2026-09-03

Responding to the latest panic over false claims that OpenAI is "doing neuralese," Dean Ball argues that technically nuanced claims are being adjudicated on X with almost no ground-truth information; communities form "thought-terminating taboos" (roon's phrase) and panic at anything vaguely resembling them, making the timeline a terrible venue for such judgment — strengthening the case for rapid institutionalization of auditing and technical assessment of frontier labs. Geoffrey Irving quotes and counters: rather than blaming people who don't know the details, labs could simply disclose the architecture to independent parties. zetalyrae amplifies the exchange.

Original post →

More from AGI Musings

AGI Musings channel →