What comes with cheap math?

According to LessWrong researcher Abram Demski, AI-assisted mathematics has exposed an unexpected bottleneck. Demski has been using Claude to develop formal proofs about when humans should trust AI systems, with Lean verification checking the work rigorously. But rigorous mathematics is no longer a signal that serious thinking happened behind it. AI can now produce impressive-looking equations without the human judgment that once guaranteed their validity. Demski frames this through economist Herbert Simon's nineteen seventy-one observation: when information became abundant, scarcity shifted from data to attention. Now, as AI makes research generation cheaper, the bottleneck shifts again—to what Demski calls care: judgment, taste, discernment. Someone still has to read the work and decide whether it actually means anything about the world. Practically speaking: Demski is publishing raw AI conversations on LessWrong for community validation, but validation is slow. He's paraphrasing each result as though reading a paper, because Lean verification doesn't guarantee the model captures reality. The irony cuts deep. AI accelerated research generation by orders of magnitude. It just relocated the bottleneck from computation to human judgment.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton