Beginning a Career in the Age of AI Discovery

Yesterday, OpenAI announced that one of its internal reasoning models had disproved a long-standing conjecture in discrete geometry.

The problem was first posed by Paul Erdős in 1946. It asks a simple question: if we place n points on a flat plane, how many pairs of points can be exactly one unit apart?

For decades, square-grid constructions were believed to be essentially optimal. Erdős himself conjectured that one could not do meaningfully better.

The model found otherwise.

It produced a new construction with a fixed polynomial improvement. Not a small numerical gain, but a real mathematical gap.

What makes the result especially striking is the method. The proof uses ideas from algebraic number theory, including class field towers and Golod–Shafarevich theory. These tools come from a very different part of mathematics from the original geometry problem.

That is the most interesting part.

The model did not simply search faster within an existing path. It connected distant areas of mathematics and found a route that human experts had not prioritised.

This is why the result feels important beyond the problem itself.

For a long time, AI has been discussed mainly as a productivity tool. It can help people write code, summarise documents, draft emails, generate images, and explain difficult concepts.

Those uses matter. But they are still mostly about making existing work faster.

This result points to something different: AI as a tool for discovery.

That does not mean human expertise becomes less important. In fact, it may mean the opposite. The proof still needed to be checked by mathematicians. The problem still needed to be understood. The result still needed to be placed in the context of decades of mathematical work.

AI can generate new ideas, but humans still need to understand, verify, interpret, and apply them.

That balance is probably where the future becomes most interesting.

As someone with a mathematics background, I found this result especially moving. Mathematics often advances through unexpected bridges: one field suddenly becomes useful in another, and a problem that looked stuck begins to open.

A geometry question solved with algebraic number theory is exactly that kind of bridge.

The surprising part is not only that AI helped with the proof. It is that the model found such a bridge.

This also feels relevant for people starting their careers now.

Jensen Huang recently told graduates that his career began at the start of the PC revolution, while ours begins at the start of the AI revolution. That comparison feels more concrete after news like this.

For students and early-career professionals, the question is not simply whether AI will replace tasks. It is also what kind of work becomes possible when reasoning tools become more powerful.

In actuarial science, insurance, and risk modelling, many problems are full of uncertainty. We work with limited data, imperfect models, rare events, and decisions that must still be made. Better reasoning tools could change how we approach pricing, capital modelling, catastrophe risk, climate risk, and financial uncertainty.

But the lesson is not to rely on AI blindly.

The lesson is to build stronger foundations.

Mathematics still matters. Statistics still matters. Programming still matters. Domain knowledge still matters. If AI becomes more capable, the value of asking good questions and judging answers carefully becomes even higher.

That is the part I find most exciting.

Not AI as a shortcut.

AI as a partner in thinking.

A way to explore more ideas, test more connections, and perhaps find paths that would otherwise be missed.

I am still early in my career. So this moment feels both challenging and motivating. The tools are changing quickly. The expectations are rising. But the opportunity is real.

Starting a career at the beginning of the AI revolution is not simple.

But the timing could not be more interesting and exciting.