The problem with proprietary models is that you don't get to poke inside and see what it is doing and how it arrives at its answer, which is precisely what mathematicians do. Mathematical understanding derives less from any particular result than the insights and methods that pave the road to results. Whenever a theorem is proved, researchers seek to unpack the proof and get inside the author's mind to learn their ways of thinking.
LLM generated results might benefit mathematical understanding if people can inspect their intermediate reasoning traces to discover erroneous human biases or patterns that they might have previously overlooked. Otherwise, the results might as well be produced by oracles.
Also, in math most "intermediate reasoning traces" are erased and the solution only shows a simplified path that many times is only visible after the proof is complete.
> The problem with proprietary models is that you don't get to poke inside and see what it is doing and how it arrives at its answer, which is precisely what mathematicians do
Uh, a hypothetical fully open source model would have the exact same problem, because LLMs rely on emergent phenomena and no one understands why they work.
>Otherwise, the results might as well be produced by oracles.
no. The math result is a result only when it includes proof. The proof is the value here. The way somebody came to it isn't really important - we don't know how Newton came to his results, whether it was apple or pear, and it isn't really important. Or how Einstein was walking the city streets looking at the tower watches - it is just historic curiosity having no real value for science.
That has been one of the greatest thing about math departments - smooth talkers were always clearly visible as smooth talkers. You're either producing proofs, or you're anything but a mathematician.
I feel for mathematicians. They have similar situation like we have in programming. Well, we all just have to evolve and adjust (in particular reign in our pride as just in a few years - i think once LLMs start hitting 100T+ - we may loose our "top of God's creation" position). Any attempts at gatekeeping, ludditing, organizing in quasi observational/advisory boards really intended to protect their tenures, etc. ... - well, you just can't stop the wave.
It all reminds how Catholic Church insisted on responsible release of the Bible in German. The Church even unleashed the devastating 30 Years War trying to protect its monopoly on religion including the right to sell indulgences, etc.
>AI labs should provide significant support, including funding
And now all those "responsible math" and advisory boards would like to preserve their monopoly on math and would like to sell the indulgences to the AI labs. As usually it is all about money and power, not about science. As a Math PhD dropout myself i feel a bit of a shame and disappointment for that undignified scramble by the mathematics establishment. Being smart they should have led the way and show an example to the rest of humanity ...
When I was a postdoc in genomics, one of my supervisors had a background in mathematics. Current bioinformatics training pipeline didn't exist yet, so most PhD students and postdocs had a background in something like CS, mathematics, statistics, or physics.
From the supervisor's perspective, people coming from pure mathematics were good at thinking about definitions. Coming up with useful definitions was the primary value they created, while theorems and proofs were just technical stuff they did to evaluate the value of proposed definitions.
My own background was in theoretical computer science, specifically algorithms. When you do algorithms without any qualifiers, you are studing them as mathematical objects in a simplified model of computation. The process often starts with a promising algorithmic idea. But if it looks like you can't prove anything nontrivial about the idea, you often stop studying it, regardless of the actual value of the idea. And if you manage to prove something, you start optimizing the algorithm for your theoretical model in order to prove better results. That usually makes it worse in practice.
The end result is that algorithms papers, both good and bad, typically contain theorems and proofs about algorithms nobody cares about. If you are a practicioner, you need to dig through all that noise to find the core algorithmic ideas, so that you can evaluate them in a more realistic setting. And if you are a theoretician, you are probably more interested in the techniques used in the proofs (which may also inspire future algorithmic ideas) than in the actual results.
You can find plenty of other similar situations. The value mathematicians create is rarely in the theorems and the proofs.
I was talking about the value created by mathematics. Which largely comes from training people to think about technical details. Which typically manifests as new ideas based on deep technical understanding of earlier ideas.
LLM generated results might benefit mathematical understanding if people can inspect their intermediate reasoning traces to discover erroneous human biases or patterns that they might have previously overlooked. Otherwise, the results might as well be produced by oracles.