Mathematicians may be worried, but AI-for-science is going to be great, recursively self-improving, and we’re going to learn loads

Simon DeDeo, a computational researcher contributing to LessWrong, gave Claude AI—a frontier model running in high-performance research mode—access to three of his own papers and asked it to extend them in novel directions. In the resulting analyses, the model traced reading patterns through thirty-five libraries across two centuries, examined parliamentary procedure across two dozen countries, and investigated how AI constructs mathematical proofs compared to humans. All three investigations produced work that DeDeo says meets publication standards and replicated his original findings while identifying new research directions. As some mathematicians express concern about displacement, DeDeo frames this as evidence that AI-for-science is delivering genuine advances—not just speed, but the ability to surface interesting research questions worth pursuing.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton