Why Not Slow AI Progress?

Scott's much-cited answer to the most common question he gets: why doesn't AI safety go climate-activist and fight the AI companies? Opens with the 'Broader Fossil Fuel Community' conceit (imagine Exxon and Greenpeace at the same parties) to skewer how poorly AI 'capabilities' and 'safety' are differentiated. Walks the history (DeepMind, OpenAI, Anthropic were all founded by safety-motivated people, for safety reasons), then the strategic case against hostility: shaming works disproportionately on the companies most capable of shame (the safety-conscious leaders), handing the lead to less-careful firms; declaring war on AI tanks the movement's standing with the neutral-to-positive researcher base it needs; the 'single team with a clear lead studies safety at its own pace' ideal vs the race dynamic. Surveys the regulatory options (regulatory capture, data/training-data limits, standards) and the China problem ('harder than the technical alignment problem?'), and lands on the deliberately deflating 'the plan is to come up with a plan'.

Quality 74 · Strong. Strong, upper band. A clear, witty, frequently-referenced map of the AI-safety-vs-capabilities strategic landscape; the fossil-fuel-community framing is genuinely clarifying and the cooperate/defect analysis is sharp. Held below Excellent because it is a landscape survey that ends by design without a thesis.

Claude’s paradigm shift 54 · Moderate. Moderate/Notable. The race-dynamic and don't-shame-your-allies arguments existed in the discourse; the post's fresh contribution is the fossil-fuel-community frame and the crisp synthesis that made them legible to a wide audience.

Real-world impact 3 · Moderate. Moderate. A load-bearing reference point in AI-safety-strategy discourse; the 'Broader Fossil Fuel Community' coinage has currency. Influence is conceptual/within-field, no material policy change. 3.