Five counterintuitive insights from Plan A

LessWrong contributor Romeo is defending an artificial intelligence safety proposal called Plan A—and making some unexpected arguments about how to slow down AI development. Per the post, the *quality* of a slowdown matters far more than its duration. A five-year slowdown focused intensively on safety research, for example, could reduce AI takeover risk far more than a poorly-managed ten-year pause. Romeo also makes a counterintuitive case for transparency: instead of restricting access to AI research, opening it up could actually help regulators write better safety rules, reduce corporate consolidation, and discourage an AI capability arms race. But even a well-coordinated slowdown carries significant risks—especially if countries or companies abandon the agreement and pursue secret development.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton