▲ 455 points
back
133 comments
Building visualizations with LLMs has been a major boost for my CS classes:
https://htmx.org/essays/universities-and-ai/#demos-visualiza...
Many visualizations that I have always wanted but just didn't have the time to build, I now have.
To give an example, I wanted a simplified 8-bit computer to complement the 16-bit teaching computer I use and designed this in a few days with the help of claude:
Using LLMs to build out the nice-to-haves that I’ve always wanted but never had time for is one of their great use cases. Visualizations are a perfect use case for this because they don’t have to be perfectly architected, maintainable code. Getting to the correct visual output is good enough, and LLMs excel at iterating something until it looks right.
I've been using LLMs to create visualizations for math papers I come across. Prompting "Create a visualization for each segment of this article in the style of a 3 brown 1 blue video using manim." has yielded impressive results.
It helps me digest the content faster and allows me to read more articles than I otherwise would.
LLMs to create and revise PIL (python image library) commands/params have saved me HOURs.
This is v cool.
When I did my microcontroller class with lecturer hand drawing an 8-bit computer, the registers, memory, instructions on the white board, it was v cool to understand how things worked under the hood.
Wondered if someone could make more simulations for what was being taught. Teaching is about deciphering a thing into it's components and seeing how they interact. Vibe coded simulations are a great tool for that.
Regarding the changes to your grading weights: https://acbart.github.io/2026/04/19/proctored-grades/
Sounds like 50/50 for the distribution? That means you are okay with a student getting a 40% across all your quizzes and then passing the class with a C-?
Terry Tao using coding agents to build apps means we're one step away from a Fields Medalist asking an LLM why his Docker container won't start, just like the rest of us.
Before LLM there has already been Fields medalist[0] who creates professional software[1].
I'm waiting for the reverse, coding agents asking Terry Tao if the proof they plan working on is worthy of a Fields Medal
This is a very humbling thought, thank you.
We thought it would level us all up to equal the best of us, but instead it has leveled them down to within our reach.
There is infinite latent demand for software, most especially outside the traditionally software-focused spaces. If LLMs stopped improving today it would take us 10 years to catch up to the new software-writing abilities that have become available. This is a great illustration of that fact.
Nice balanced perspective there at the end:
"as such [LLM-coded interactive] supplements are not mission-critical to the core of the paper, I again feel that the downside risk of using guided interaction with LLM agents to generate such visualizations is acceptable."
It's a tool. Good for some things but not others and generally not to be trusted.
> It’s a tool. Good for some things but not for others and generally not to be trusted.
I agree completely you always need to check the work of LLM agents, but it does strike me as a tiny bit funny to anthropomorphize AI by using ‘trust’ while warning against anthropomorphizing the AI by using unchecked output. ;) Generally speaking, “trust” in AI has been going up very quickly as the models & harnesses improve, and as people figure out effective workflows.
I trust my hammer with nails but not screws… does that mean the hammer should generally not be trusted? The problem with AI is we don’t know the difference between nails and screws. (This may be where my analogy breaks down. :P) But I feel like saying don’t trust it isn’t as helpful as saying something like you should expect to spend more time planning and iterating than before, and you should expect tot spend more time reviewing and checking output than before, and learn how to use skills and context and subagents, and learn to use AI on some non-production low-consequence projects first. Saying ‘generally not to be trusted’ implicitly suggests not using AI, and doesn’t leave the reader with how to use AI. The goal is to build trust by building good workflows and by understanding what works well and what doesn’t, right?
I don't understand what trust means in this context. Even if I were able to hire Donald Knuth to write all my code, I wouldn't "trust" it to be bug-free, let alone to be the right fit for my needs.
> and generally not to be trusted
There are many AI bulls who adamantly disagree and cite Tao’s statements about LLMs for mathematical proofs as an example of how advanced and autonomous these systems already are
Indeed. LLMs produce truly atrocious code, unmaintainable and unreliable. If you're vibecoding a toy to amuse yourself or something similar low-stakes, that's perfectly fine! For higher-stakes code, it's definitely not.
Terry Tao using coding agents feels like watching a Michelin-starred chef discover microwave dinners and get genuinely excited about them.
I liked this article about an old recipe book and what cooking could have looked like if we took microwave cooking seriously: https://malmesbury.substack.com/p/my-journey-to-the-microwav...
Running legacy educational Java applets, especially around math and physics, has been a longstanding popular use case of our CheerpJ Applet Runner extension, running Java bytecode in the browser via WebAssembly.
I am not sure how to feel about agents solving the problem via proper modernization. It's certainly positive that students will be able to interact with this content in a modern and more accessible way, but the educational use case for our product, although not commercially important, has always been a source of pride.
https://chromewebstore.google.com/detail/cheerpj-applet-runn...
I always enjoy these "domain expert has fun using AI to do something in their domain" articles. But it's always a hobby project, never something serious.
This post inspired me to have Claude port my 30 year old high school German Java applet game to Javascript, complete with a faked git history:
https://github.com/bradfitz/koffer#der-verloren-koffe
Play online at https://bradfitz.github.io/koffer/js/
So neat seeing ~30 year old code come back alive.
The article's awkward opening statement proves it wasn't written by AI.
I have been interested in machine-assisted ways to do and teach mathematics from as far back as 1999, when I started coding several applets in Java 1.0, both for my complex analysis and linear algebra courses, to visualize various mathematical objects I was interested in (such as honeycombs or Besicovitch sets).
I laughed at this "but the code complexity became too much for me, and I abandoned the project." even Terry Tao finds some code too complex to write.
Really bullish on LLMs expanding code development by a very large group of people who are really smart in some domain but could not get into 'coding'.
Using LLMs to generate dashboards is probably their most productive use case
It's probably a matter of short time until it's possible to disassemble any sophisticated software, rewrite it entirely with better features and usability, generate all needed artifacts, port to any platform. The only moat left is probably remote massive data storage. So if you want to replicate YouTube or TikTok, it's not impossible, but requires a lot more hardware assets than say anything that runs entirely locally (like operating systems or most video games).
I am far from a mathematician but I am excited by the possibilities of using AI for generating more math. Math in my mind exists purely in the world of forms, and cannot be appropriated for profit, but is downstream to everything else. I am keen to see what this enables.
even though there's still a lot of work to push things over the finish line, i have enjoyed how much it has reduced the activation energy for starting and finishing "one of these days..." projects!
I wonder if LLMs modify their output when they realize they are interacting with a famous person.
By famous I mean someone whose biography is in the training data. All models know a lot more about Terrance Tao than they know about me, when he's working on his projects do the models know they don't need to explain "Besicovitch sets".
Since the system prompt likely includes something about not insulting the user, does the LLM modify it's responses if it realizes it's talking to famous politician, like "dont mention the time $politician was cancelled".
His website using mathematical knowledge is refreshing. There's a small UI bug, but personally, I wish more educational materials were this rich in audiovisual content.
Interesting perspective. One thing I'm curious about is whether modern coding agents are actually better at maintaining older codebases than greenfield projects. In my experience, legacy code often lacks the context that agents need.
Using Fable with Anthropic Design skill works every time.
This is amazing!
JavaScript: Good Enough for Terry Tao
Nice write-up!
Nice
LLM will do very good job in pure mathematics since it don't need the senses to logically understand/conclude a given topic.
The more Terry talks about AI, the more I'm starting to feel like Terry may have some undisclosed conflicts of interest.
https://www.reddit.com/r/mathematics/comments/1tryyw7/terenc...