Ten advances in mathematics and theoretical computer science

OpenAI dropped a rundown this week of ten problems it says its systems helped resolve, spanning geometry, cryptography, and computational complexity. These aren't toy puzzles. They're the kind of questions that mathematicians and theorists have chipped away at for years, sometimes decades, without a clean resolution. The framing from OpenAI is straightforward: this is a demonstration that its models can now function as genuine collaborators on frontier theoretical work, not just tools for summarizing papers or checking homework.

What makes this batch interesting is the spread. Geometry problems tend to be about structure — how shapes, spaces, and objects relate to each other under strict rules. Cryptography research often hinges on proving that certain problems are hard enough that breaking an encryption scheme would require unrealistic computing power. Complexity theory sits underneath both, asking how much time or resource any algorithm needs to solve a given problem, and whether some problems are fundamentally harder than others. Progress across all three at once suggests something more general than a lucky pattern-match on a single subfield.

Still, the source material here is thin on the details that mathematicians will actually care about. There's no walkthrough of proof techniques, no named theorems, no indication of which specific conjectures moved from open to closed. That matters because in pure math, a claimed result isn't real until it survives scrutiny from people who didn't write it. Peer review in this space is slow and unforgiving on purpose, and rightly so — history is full of computer-assisted proofs that needed months of independent verification before anyone trusted them.

This fits a pattern that's been building for a couple of years now, from DeepMind's geometry-solving systems to various efforts pairing large language models with formal proof checkers. The pitch is always the same: AI as a tireless research assistant that can explore vast search spaces humans can't. OpenAI's announcement reads like another entry in that same race, positioning itself as a serious contender in mathematical research rather than just chatbots and coding assistants.

Whether this becomes a genuine shift in how theoretical work gets done, or another headline that fades once outside mathematicians pick it apart, depends entirely on what OpenAI publishes next. Ten results with no proofs attached is a press release. Ten results with proofs that hold up is a different story entirely.