AI is making progress on Erdős problems; verification is the bottleneck

A solved problem still has to become knowledge

Quanta’s report describes AI-assisted progress on problems associated with Paul Erdős. The immediate story is mathematical capability: models can search broadly across techniques and persist through detail-heavy lines of reasoning.

HN’s reaction supplies the necessary second act. One reader asked how society can use breakthroughs that even career mathematicians may not understand. Another suggested that long-standing physics, chemistry, and biology problems would make the same question much more urgent.

More proofs require more checking

A particularly useful comment identified the operational bottleneck: automated systems may combine ideas from many papers and patiently explore possibilities, but a flood of proposed arguments needs automatic formalization and checking because manual review cannot scale indefinitely.

That is not a reason to discount the result. It is a reason to distinguish discovery from accepted knowledge. A candidate proof can direct human attention; a verified proof can become infrastructure for the next theorem.

The next phase of AI mathematics is therefore likely to be judged less by a count of famous problems than by the quality of its verification pipeline: can an argument be formalized, independently checked, and made legible enough for mathematicians to extend? That is where a striking answer becomes durable science.