LLMs are solving Erdős problems

The Man Who Loved Only Numbers is a biography of mathematician Paul Erdős written by Paul Hoffman.

It's a great read, the maths is accessible, and the human story is quirky. Paul was still doing maths (and forgetting to do almost everything else) until the day he died. He left thousands of problems unsolved that he never got time for: these are known as Erdős problems.

I usually post about AI, and Erdős problems are relevant here because large language models are now solving them.

This is a really big deal. Calling it a watershed would not be unjustified. To be fair, not all of humanity has been trying to solve Erdős problems, so this is not quite like solving fusion. But it is now overly simplistic to call LLMs probability machines, or "next token generators".

The criterion for a PhD (at most universities) is to add to the body of human knowledge. Solving these problems proves that LLMs are now adding to human knowledge in a real, quantifiable way.

As a CEO, it should be clear that this is a massive opportunity: PhD level talent is available at very low monthly cost; the challenge is how will you turn that into ROI?

This is not a trivial problem to solve (it may require staff training, data sanitisation or deeper technical guidance), but surely it's not as hard a problem as one of Paul's.

Maybe AI can help you solve it.