Given the recent progress in AI solving many long-standing open problems in maths, there is no doubt there are many schools of thought around the use of AI in maths academia. However, I want to provide a different perspective of the worrying role of AI (specifically LLMs) in maths communication from my 7 years of experience in running a maths YouTube channel, and perhaps this new perspective can apply to academia too.
Two kinds of maths content
In maths communication, I like to separate content (YouTube, blogs, StackExchange, or otherwise) into two main categories: traditional topics and novel topics. I define “traditional” in the sense that it appears in a typical maths curriculum in schools and universities, e.g. my videos about the Jacobian determinant, Green’s functions, and complex analysis. “Novel” topics would include a geometric interpretation of Taylor series of sin x, and the Stein’s paradox in statistics.
I dabbled in both kinds of maths content, but I strongly prefer to watch and make videos of the novel topics category. Perhaps it is because I am chronically online and have watched too many maths videos on YouTube, and I want some fresh air; but the outside-curriculum maths is what draws me into maths in the first place. I frankly would not have loved maths without knowing there are so many maths topics outside of my high school curriculum.
However, based on my YouTube analytics, most people prefer the traditional topics category (in roughly 2:1 traditional to novel). I lamented about this in another blog, but the reality is that many viewers who watch maths videos on YouTube just want a more visual and didactic explanation for concepts not taught well in lectures. Many people treat maths as a tool that they need to learn how to use, rather than an interesting subject to explore per se. I need to emphasise this point because this likely contradicts the philosophy of the readers of this blog.
When AI comes in
As AI progresses, students start using ChatGPT for their maths studies, which is heartbreaking for university tutors like me walking around in tutorials. (“You literally paid me to answer your questions, and you are using ChatGPT???”) At the same time, some people realise that you can use LLMs to generate maths videos that explain these “traditional” maths concepts. These spur the development of many Edtech websites. If your goal is to learn maths in the curriculum, even if you are a visual / audio learner, many students can do without a human in the loop.
What about videos in the “novel topics” category? A particularly popular maths channel on YouTube is 3Blue1Brown, who has an idiosyncratic visual style developed using his own Python library known as Manim. ChatGPT or other LLMs are surprisingly good at coding, so some people took this opportunity to use LLMs to mass generate videos in that visual style by coding with Manim, so YouTube was flooded with them for some time.
A lot of their output is just rehashing content made by other maths channels, notably Numberphile. I know that because, as I said, I am chronically online and watch too many maths videos; but to a lot of viewers on YouTube, they might be “novel topics”. Many people do not realise that they are AI-generated, and even if they do, they do not mind. These AI-generated content farms have therefore become quite successful.
Therefore, videos in both traditional and novel topics can be automated with LLMs nowadays. Since many people treat maths as something they need to learn rather than an art, they do not care whether it is a human or an LLM teaching them.
If the goal of maths communication is to teach maths concepts in a better way, perhaps one can still argue that these AI-generated maths videos are bad at explaining concepts clearly right now, as LLMs are often too succinct. If the goal is to expose more exotic maths topics, perhaps we can also argue that AIs are still just rehashing content that is already present on YouTube.
However, I can absolutely imagine that these will improve in the future. In some sense, it is a losing battle against AI used in maths communication, and they will be especially popular among people who take a utilitarian view of maths, which might be most people. There is always a vocal group decrying the use of AI on social media, and that includes myself, but we have to face the reality that the vast, vast majority of the general public really do not care.
A silver lining
Despite all the pessimism, I still think humans can trump AI in one thing: conveying passion. It is very difficult for a teacher to hide the enthusiasm (and the indifference) of the subject, and this passion (or apathy) is contagious. I remember that my chemistry teacher in high school was very passionate, which made me passionate about it too, searching for many cool experiments on the internet. The next year, I was taught by a different chemistry teacher who clearly did not care, and my interest in chemistry waned.
The general public already has a negative predisposition about maths being purely as a subject that is just about right and wrong. And especially given the development of AI in recent weeks, maths might be seen as more soulless than it already was. The current goal of maths communication is perhaps to humanise maths with our palpable excitement, to get the general public to care about maths from its inherent beauty rather than its usefulness, e.g. this famous video on Klein bottles on Numberphile.
It does not mean we need to exaggerate our facial expressions to convince others of our excitement on the topic, because authenticity is crucial in any form of communication. However, there needs to be a very strong motivation in the beginning that even laypeople can understand and relate to, perhaps even with personal anecdotes. This is not easy, because we have to guess what gets an outsider excited, and we have not been one for a long time.
If you told me 6 years ago that I have this reflection now, I would not have believed you, because I have been trying to avoid parasocialism for a long time, which is why I was faceless on the channel. I do not want people to follow me for myself, but for my work. Given how generative AIs are knocking on the doorsteps of maths communicators, if I want my videos to still be watched, I might have to show more of my personality – the one thing that LLMs cannot touch on. This is the inconvenient truth that I am still adapting to.
Lessons for academia?
All the discussions above apply to academia too1. Of course, there is no need to convince others in the community that you are passionate about maths (if you are not, you are in the wrong place). But why you care about your research, and most importantly why they should also care about your research, is quintessential. After all, academia is maths communication, just with a different audience.
In the most pessimistic scenario where LLMs can solve any open problem you throw at them given enough time (and tokens), why anyone cares about the solution is still up to humans to decide. Perhaps it is because this counterexample defies the intuition one might have about this problem, or maybe this proof can bridge two different areas of maths long thought to be unrelated. The story of why that problem matters is often more important than the solution.
Perhaps similar to maths communicators, academics need to adapt too: from doing research to proving theorems, to understanding the results an LLM might come up with, and more importantly explaining why they are important. Sure, AIs can solve a lot of open problems; but it cannot explain why researchers and the general public should care. We, as humans, have a much better shot at this. This could be our role in the future, at best a hybrid with research, an inconvenient truth that we might have to accept sooner or later.
A plea for academics to do outreach
From the aforementioned perspective, it makes sense to do maths outreach, especially at the level of the research that you do, to get a head start. From a practical point of view, you have to convince the policymakers and grant issuers that your work is interesting, so some form of outreach has to be done. From an ethical perspective, it also makes perfect sense to explain to taxpayers, who fund your research, what you do. Given the anti-intellectualism that drives countries, including but not limited to the US, to reduce funding on maths and basic scientific research, it even makes sense politically to do outreach.
Yes, you will have to dumb down slightly, but not necessarily to the degree you think. I am currently making a series of videos on differential forms, and still tens of thousands of people watch them. It takes time to hone the skills in doing any outreach, because it is a very different skill from research, and you will need to convince people that the content (maybe YouTube videos like me, or blog posts, or anything in between) is worth watching from the beginning, and this is something that an LLM does not do well, yet.
In fact, a hot take of mine is that doing outreach actually improves the research community as a whole. Outreach usually greatly enhances your understanding of your research – I accidentally learn a lot more than what I needed to present when crafting my script and animations in my videos. When we do more outreach, we also understand better how to communicate maths in a research paper, so papers can focus on communicating ideas rather than over-emphasising rigour and brevity, helping others in the community to understand more efficiently.
Throughout this post, I have presented a very bleak outlook, especially for mathematics research, in favour of promoting maths outreach. This is partly to prepare for the worst, but partly because however much I do not want it, the capitalist, results-matter-more-than-process ideology epitomised by these LLMs had never lost in the past. There is little to no reason why the AI companies stop developing and attempting to replace any cognitive labour, especially mathematicians who are the symbol of the intellectual workforce.
It does not mean that we should completely give up on protesting AI encroaching on research mathematics or maths communication, but we need to at least mentally prepare for the scenario that they do. What really cannot be replaced by AIs at the moment is doing outreach in a way that makes your enthusiasm infectious, and even if you still hold on to research, as you should, there is still a lot of merit in engaging in maths communication.
Received on 18 August 2026.
- I am currently a PhD student in physics, so I am not entirely sure how the maths academia reacts to all these. But the broader point still hopefully stands. ↩︎
Add to the discussion