Armin Ronacher on P(doom), pacing, and the AI commons
Ronacher reframes the AI-safety pacing debate around public-resource strain, closed-lab concentration, and the commons that model labs trained on and now tax.
Links: Original source · Shared link
Logged at IST: 2026-09-15 22:18 IST
What it is: Mario Zechner recommended Armin Ronacher's essay on P(doom), Dario Amodei's frontier-pacing argument, and the politics of closed AI labs.
Gist: Ronacher says he shares many of the same observations and worries as the frontier-lab safety crowd, but objects to the implied cure. The danger he foregrounds is not near-term human extinction; it is the damage that closed, subsidized AI systems do to everyone outside the labs: open-source infrastructure gets stressed, universities and companies pay a model-provider tax, and public-data-derived capability gets rented back from a handful of American firms.
His counterpoint to "pace the frontier" is that open weights and capability diffusion can themselves impose pacing. If powerful models are accessible on more equal terms, the economics are less distorted, the public has more ability to inspect and adapt, and the geopolitical argument for keeping capability concentrated in OpenAI/Anthropic-style labs becomes weaker.
Newsletter angle: A useful European/open-commons response to the Amodei/Altman/Musk pacing discourse: the policy question is not only whether frontier models are dangerous, but who gets to own and meter a technology trained from the commons.