AI just moved from solving textbook problems to cracking long-standing open questions in geometry, cryptography, and computational complexity — and that changes what we thought AI could do.
The Breakthroughs: What AI Actually Solved
OpenAI has published results on ten advances across mathematics and theoretical computer science — fields where problems can sit unsolved for decades. The areas include geometry, cryptography, and complexity theory, each of which underpins how we build secure systems, design algorithms, and understand the fundamental limits of computation.
These aren't incremental improvements on known techniques. Open problems in these fields are the kind that entire PhD careers get built around. The fact that an AI system is contributing meaningful progress — not just verifying known proofs — marks a genuine shift in what large language models can do when pointed at hard, abstract reasoning.
Why Cryptography and Complexity Theory Matter Most Here
Of the three domains, cryptography and complexity theory carry the most immediate real-world weight. Cryptography secures every password, payment, and private message you send. Complexity theory determines whether problems are solvable at all — and how fast. Advances here don't stay academic for long.
If AI can now assist in finding new results in these areas, it becomes a genuine research accelerator — not just a coding assistant or summarisation tool. Think of it as the difference between AI helping you write an essay and AI helping you discover something new. This is the latter.
For a deeper look at how AI reasoning capabilities are evolving toward this kind of frontier thinking, the course The AGI Race puts these developments in sharp context.
What This Means for Learners
You don't need a maths PhD to take something useful from this story. What it signals is that AI's value is rapidly shifting from retrieval (finding things it's seen before) to reasoning (working through things it hasn't). That changes how you should be using AI tools right now.
Prompting AI to reason step-by-step through hard problems — rather than just asking for answers — is a skill worth building today. Understanding how language models actually process and generate structured reasoning is the foundation of that skill, and How Neural Networks Really Work is the place to start.
The practical takeaway: AI is becoming a thinking partner for expert-level problems. The people who know how to direct that thinking will have an enormous edge.
