Hacker News Reader

Comment by OutOfHere | Hacker News Reader

The obvious answer is to have mathematicians use AI to:

1. Help understand, check, and explain the results.

2. Write new works explaining or refuting the new approaches and results in more lucid language.

3. Advance the field further.

I don't know why this is not obvious. Each step is intended to support human understanding, not to replace it. Any mathematicians who don't do these will be left behind, and if none do it, the field of human mathematics itself will become obsolete. All I am hearing so far is excuses.

View on HN

AI is producing works that are so poorly written/explained that it requires AI to even parse the results, which in turn produces explanations that are still confusing. In fact, it seems the humans are required to understand and explain the results, and the fact that they need to use AI to do so is a shortcoming of AI.

And if humans decide to give up on mathematics because the process of using the machine is so tedious, then there will be no value in automated theorem proving.

I am sorry but your resistance will fail utterly as long as AI itself can understand prior AI works. This is the only requirement that need be met. If it helps, imagine that the work is by a culturally distinct alien race in an alien language, and if it isn't there now, it's might get there soon.

Check this:

> resistance will fail utterly as long as AI itself can understand prior AI works

That's not the issue in the long term. Look further: where did these questions come from? They came from humans who had been exploring mathematics and noticed interesting patterns that cropped up again and again.

The people who posed these questions had a profound understanding of the background maths that led to them. Me - I don't. Take one of the results he quotes: "Positive solution to the Unitary Synthesis Problem". He may as well be one of your alien races yabbering to me in an alien language. The question is meaningless to me. The answer the AI came up with, correct though it may be, is equally meaningless. You may as well have told my dog how to find eigenvalues.

If we let AIs do all the understanding so we know nothing about math, there will be no human understanding that lets us find new questions, or understand the value of any questions AI might pose and find the answers to.

Progress in science assumes a certain level of signal to noise ratio in papers. Decreasing the signal to noise ratio in published results will eventually overwhelm the ability to successfully disseminate newly obtained knowledge. This is already happening in the open source community with maintainers being pushed to the point of burnout by AI slop. The same thing is now beginning to happen in other fields.

This is only going to get worse in the foreseeable future.

You can certainly do that - but it's quite the break from tradition to release a paper in the state described. Why they did is a really interesting question! It may be that AI math requires approaches that humans don't find intuitive and what you are describing is actually counter productive (because, in summarizing the work in a way humans understand, you're removing the context an AI would use to further the work an AI did). It also might be that OpenAI could have done that and chose not to - or maybe they tried and this was the best they could do. No matter what I don't think anything about how to react to a paper being released in this state is obvious.

> in summarizing the work in a way humans understand, you're removing the context an AI would use to further the work an AI did

The "summarization" process can be multi-step. Initial steps develop new frameworks and prerequisite concepts. Later steps build upon these frameworks and concepts. Some can be simplified too if this can be done without loss of fidelity. The summarization comes last, and is meant to preserve sufficient context.

> I don't think anything about how to react to a paper being released in this state is obvious.

If it helps, exaggerate the condition a bit by imagining coming across a repository of alien knowledge.

> Some can be simplified too if this can be done without loss of fidelity.

If you don't understand why the paper is in its current form, how do you know what changes would damage the fidelity for the AI? I think your example of alien knowledge is misleading because it suggests the aliens don't need to understand the info - but in this case we intend to keep using AIs to expand this work. So if you transform it into something that humans think makes more sense without understanding why it's in a form that we hate now - you risk losing something in the form that you don't understand.

In certain fields if you are talented and just starting out, you have better do any work that’s important to you offline for privacy/pirate reasons I really can’t say that enough, those AI model companies are not even remotely your friend.

That makes zero sense to me since it's effective use of AI that can propel one's career ahead.

Isn't AI well-suited to tasks #1 and #2, though?

#3 at this point might need more human intuition; but that might be a 2026 problem.

Reply on news.ycombinator.com