Asterisk/Zvi on California's AI Bill

A timely explainer of California's SB1047 AI bill (via Zvi's Asterisk piece + FAQ) that cuts through widespread misrepresentation: it applies only to >10^26-FLOP frontier models, doesn't ban open source, doesn't burden small startups, the perjury clause is routine, and the testing adds <1% to training cost. Covers the reasonable objections (the pre-training-knowledge clause, derivative-model edge cases, benchmark vagueness) and ties back to the Origins-of-Woke 'laws get reinterpreted' concern. Endorses the bill alongside Bengio/Hinton.

Quality 69 · Strong. Strong: a clear, useful policy explainer that corrects real misinformation on a live legislative fight; topical and somewhat derivative of Zvi's analysis.

Claude’s paradigm shift 48 · Moderate. Notable: the who's-honest-check framing + the myth-by-myth debunking, building on Zvi's work.

Real-world impact 2 · Minor. Fed into the live SB1047 discourse (Scott endorsing it publicly); explanatory rather than institution-moving. =2.