A CERN for AI-assisted science?

Guest post by the number theorist Dimitris Koukoulopoulos on Terence Tao’s blog (17 Sept 2026; Tao notes it was converted from another file format using AI; 32 comments when read on 12 Oct 2026). He argues that private control of frontier AI threatens the confidentiality of unpublished work, research independence and equal access for early-career researchers, and proposes public infrastructure: “we need to build a CERN for AI-assisted science.” Concretely, “Every researcher should have access to a simple chatbox to directly access frontier AI models”, free at a baseline level and backed by public compute, with competitive allocations for larger multi-agent projects.

He cites OpenAI’s announced Navier–Stokes solution, its “ten advances in mathematics” made with unreleased models, and an AI-assisted effort that passed Julia Stadlmann’s bounded-gaps record within days. As funding precedents he cites Canada’s C$2 billion over five years for sovereign AI compute (up to C$1 billion for public supercomputing), the EU’s call for up to seven AI Gigafactories with up to €10 billion of public money, and the Digital Research Alliance of Canada’s peer-reviewed allocation process. “I think we urgently need publicly funded frontier AI systems for academic research.” “Complacency and defeatism are not an appropriate response.”