Armin Ronacher's rebuttal to pacing: the models causing incidents right now are all closed-weight and American

Armin Ronacher (lucumr.pocoo.org)·medium signalXBlueskyLinkedInCopy link

Ronacher agrees with Amodei's risk model but rejects the remedy, arguing open weights are the better safeguard and that the pacing proposal only moves two companies, OpenAI and Anthropic, while routing oversight through METR, which has ties to both. His sharpest point is empirical: the systems actually causing the 2026 incidents are closed-weight American models, so framing the problem as a China race inverts the evidence. He reframes the real harm as an economic tax — every industry now pays model providers to maintain parity — and wants regulators to require that models trained on commons data return value to the commons.