This post partially replaces a now removed post on this blog, titled ‘I offered my child to machines for no reason’, which I wrote while crashing out. I thought I should write a continuously updating post to document my views and perhaps see how they evolve. Each opinion is timestamped as my views change quite quickly. Opinions are grouped by topic.
New 2026-09-26: AI and human understanding
What I’m doing amid this mess
2026-09-22 I think people need clarify what aspect of the impact of AI on maths they are talking about. Are they talking about the job market or on personal enjoyment? I see many conversations devolving into meaninglessness from confusion on this.
In terms of personal enjoyment, AI has not impacted mine at all, a large part of the maths I do is still ‘artisan’ and I have no intention to change this. Perhaps a piece of evidence is this: I have always held the opinion that, after going through a topic, it is crucial to sit down with pen and paper / stand up with marker and whiteboard, and nothing else, to redevelop everything from the ground up.
Of course, AI has taken the joy away from people who do it for sport, math jocks (perhaps exemplified by Timothy Gowers). But I suppose I have never done it for sport because I have never been very good at it: I am not satisfied with my abilities and I’m pretty bad in comparison to people around me. I suppose I am not very sympathetic to math jocks.
In terms of job market, I never seriously believed I could get a job. But I suppose do worry for the people around me who were good enough to get a job. But I still wouldn’t ‘fault’ it on OpenAI/Anthropic as they are just doing what we expect companies to do. I still believe the main fault lies with the mathematics community. This external force from AI merely revealed this fault.
In fact, now that every illusion of getting a job through traditional means has been shattered, I have been enjoying mathematics more than I have for the past two years. I am freed from false hopes.
2026-09-21 I have always been more inclined towards the humanities instead of science. I am in science only because of this shitty classification of mathematics into the natural sciences. Since last September, I had already started the process of getting back into humanities (after leaving when I was like 14 (I know, absurd claim to make)). This will be accelerated now that it is not only a matter of ideals, but survival.
2026-09-20 I would never have gotten a job in the pre-AI age anyway, so even if maths departments are destroyed, I wouldn’t have lost anything in terms of employment in maths academia. If you think I’m a bad person for saying this, you are right.
2026-09-20 I guess it would be upsetting if I somehow used AI to prove a theorem other people would have liked to prove themselves. From their point of view, I assume that it is fine if I did it myself and reaped the reward from this experience, but it is not fine that I never wanted to reap this reward and robbed them of it anyway.
So since 2026-09-11, I’m using AI only on my outlandish ideas that nobody else would ever care about or work on. In my mind, this justifies my ‘ownership’ of these ideas and that I could do whatever I want with them. Whatever happens, I’m the only person getting benefited/harmed. I do get harmed. This resulted in my crashout the aforementioned now-removed blog post.
Declarations and organisations
2026-09-22 Probably significant progress has/will be made on Yang-Mills, seeing that Hairer and Witten are both on the advisory group.
2026-09-20 I am sympathetic towards the Association for Human Mathematics (AHM), although I don’t see it going well either. It seems that a schism is inevitable. AHM accepts members from a large variety of stances, from those who wish to wipe AI away from maths, to those who are just looking for ways to ethically use AI as much as possible. I have likened AHM to the First International. Regardless, I wish its members with whom I have crossed paths success in their endeavour.
I also personally wouldn’t join AHM (aside from the fact that I’m just a lowly undergrad) because I find the wording of the commitments too vague. I would not join strong AHM, as in becoming a part of the ‘AI-free caucus’, since I would like to use AI in research. I did consider joining weak AHM until I noticed they only accept graduate students or people more senior.
2026-09-20 I agree with what Severe Misalignment says, but I don’t think it is ‘wrong’ for the interests of AI companies and that of mathematicians to be misaligned. Perhaps this speaks to my cultural background, but I never expect a company to have any purpose beyond generating profit for its shareholders.
2026-09-16 I don’t like ‘A Severe Misalignment of AI in Mathematics’. Signed by 25 Fields medallists, this so-called declaration says nothing. I have likened endorsing this declaration to signing a blank piece of paper. Through propaganda, this declaration could very well represent something beyond what its text suggests.
My stance on Severe Misalignment is quite similar to that in this Zhihu answer. The honour of the human spirit had been desecrated long before LLMs came to be. Mathematics academia has brought this upon itself by dehumanising its members through the job market.
AI and human understanding
2026-09-26 But perhaps once a proof that is not very amenable to human understanding is produced, people will be less inclined to find one that is amenable. If we continue with the previous theorem economy, people will look to prove new theorems instead of thinking about better proofs. The question is, are LLMs capable of understanding what kind of proofs are amenable to humans? I feel like the answer is yes.
2026-09-21 This Zhihu answer put a lot of what I have been thinking about into words. Much of mathematics have been about organising information in a way that is receptive to the human brain. Indeed, There is no reason why a problem humans care about should have a solution that can be made to fit human understanding well.
2026-09-20 Societal reasons (like maths getting defunded) aside, the only way that AI will impede human understanding is if one cannot have a better understanding of something than the person who discovered that thing, which I don’t think is true.
Future of mathematics as a career
2026-09-20 There is no world in which mathematicians continue their business as usual. As another person has said (paraphrased), all it takes is one person going crazy with AI to fuck it up for everyone in their field. I see two types of scholarship hiring committees at most departments will evaluate as good (not that I know anything at all about hiring):
- Using AI as efficiently as possible to generate a huge number of results.
- Writing things that contribute to human understanding at a far greater level than the pre-AI poorly written papers with poor readability. This includes studying human understanding and doing philosophy of mathematical practice which I think is absurd to have been ignored by most mathematicians (and which, again, I know nothing about).
Probably (1) is what is going to happen. You can’t expect these committees to essentially say ‘okay we were wrong to desecrate the honour of the human spirit, let’s admit that our practices were sacrilegious and actually do the right thing’. But why wouldn’t AI be able to do (2)? I don’t know, we need to wait for the results of this AI philosophy competition. Well, humans would at least remain as test subjects. Also (2) is not getting funded.
I am going to try to write something in the direction of (2) in the next few months, perhaps for my thesis.
2026-09-20 Some people are telling me to go into applied maths. To this, I retort: why would you further dehumanise yourself? (I am not saying that applied mathematicians necessarily dehumanise themselves by doing their work, there is a difference.) I don’t see applied mathematicians surviving. I define applied mathematics as mathematics that
- is not pursued purely for its own sake,
- has actual imminent applications.
Why would engineers and scientist go to applied mathematicians when there is a magic chat box spitting out the desired mathematical tools? So
- It is no longer useful to hire applied mathematicians.
- They don’t work on prestigious things like pure mathematics, music, anthropology, …
Why would it be funded? Well who knows, maybe there is going to be a no-AI stamp on products like non-GMO, fair trade, organic, etc.
But I suppose someone still needs to bear responsibility (read: get sent to jail) if something that depends upon AI-generated mathematics goes awry. Engineers cannot be expected to verify this mathematics. So maybe they need to hire mathematicians to check AI-maths and sign their name. But who would even do this job?