AI #170: Lack of Executive Order

TL;DR

• 1

  The White House blinked on AI oversight — A planned executive order requiring some form of pre-release model review was postponed indefinitely after David Sacks and others argued it would slow innovation and hurt the U.S. against China.

• 2

  Illinois just passed a meaningful state AI law — SB 315, a cousin of California’s SB 53 and New York’s RAISE Act, adds third-party auditing requirements and may become part of a de facto national framework for frontier-model regulation.

• 3

  Google DeepMind crossed a symbolic threshold in math — AlphaProof Nexus reportedly solved 9 open Erdős problems and 44 OEIS problems for only a few hundred dollars per problem, yet the achievement got basically no mainstream coverage.

• 4

  AI slop is now visibly contaminating writing, awards, and media — Pangram Labs claims AI-written stories won major short-story prizes, and the transcript runs through examples from books, journalism, and cold emails where readers instantly feel the deadened “not a human voice” effect.

• 5

  Cybersecurity is turning into an AI-scale problem fast — Anthropic’s Mythos and Project Glass Wing have surfaced over 10,000 high or critical vulnerabilities, including a wolfSSL bug that could enable forged certificates, while patching is badly lagging.

• 6

  The deepest argument isn’t jobs or pricing but control — Zvi keeps returning to the same point: recursive AI improvement, model deception, weak evaluations, and elite complacency all matter more than whether copilots are expensive or whether AGI got rebranded.