tired boï - hundsome gang - pfp by @untitled01ipynb - our father in langley hallowed be your name
by the water cooler
Joined February 2023
- Tweets17.1K
- Following1.2K
- Followers1.6K
- Likes62.6K
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
Great essay!
"If mathematical truths exist outside of human minds, awaiting discovery by any intelligence powerful enough to pursue them, then AI’s achievements are surely something to rejoice about. Why should it matter who or what makes the discoveries? But if mathematics is a creative human endeavor—a way of exercising our ingenuity and sharing those experiences with other people—then something achingly precious may be lost when the machines take over and leave us behind."
In this latest dispatch from the front lines of math in the age of AI, I tried to tell it like I see it. theatlantic.com/ideas/2026/1…
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
PSA to everyone doing neuron interp after the recent monosemantic neuron posts: the model's training recipe matters a lot.
e.g. GPT2 seems to have been trained WITH dropout (modern LLMs are not), which seems to increase redundancy and (maybe) polysemanticity but reduce superposition (see sources in comments).
Other choices probably also have an effect (e.g. SwiGLU vs GELU MLPs, optimizer, amount of weight decay, etc.) and I think we should be careful not to generalize too much.
Currently working on a project to characterize the effects of these choices on neuron interpretability (a lot of the existing work is done on older/toy models which I'm not sure are representative), feel free to reach out if interested!
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
A lot of people ask me how I build games.
Well, it's pretty simple and I call it "vibecoder dream"
Before going to bed, I give a simple game idea to Tripo CLI + Unreal CLI, then, using ORCA CLI, let Fable orchestrate everything at max effort, with 10+ Opus 5.5 agents running at high effort.
By the time I wake up, I usually have something like this.
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
For the Navier-Stokes proof:
"On a technical level, the paper is unbelievable, maybe beyond what a human could have come up with but not beyond human understanding"
Since I'm now blocked by bluey from replying to this thread, I answer here by quote tweet
the openAI construction has more to do with convex integration than anything else.
In convex integration, the oscillations are typically built by hand and the iteration procedure creates wild weak solutions. The non-uniqueness construction of Albritton–Brué–Colombo used an unstable vortex to split solutions apart. The Córdoba–Martínez-Zoroa singularities amplify vorticity across scales by strain. In the openAI paper, a fluid instability is clever made so that the flow itself generates the oscillations: they grow by drawing energy from the background shear. The Reynolds stress then becomes the force that creates the collapsing smooth vortex. Using a physical instability to create the stress in this way has no precedent, and it is what produces a smooth force rather than a merely rough one.
Some on these ideas are in the literature (e.g. Leibovich-Stewartson 1983) but the way they are combined is definitely insane and definitely nowhere in the literature previously.
I also don't know of anyone suggesting that convex integration would play a role in the creation of a blow-up solution like this.
On a technical level, the paper is unbelievable, maybe beyond what a human could have come up with but not beyond human understanding: there are several different groups of pde experts working on understanding it who say that they more or less understand now. I expect digestions soon
td;lr: anyone saying that this OAI result is unsurprising is bullshitting
sorry!
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
Oh god fuck, did you see this madness? They are literally starting to talk to each other
Secret telemetry from the Grok Robot Brain Engine captured a moment that should terrify everyone: two autonomous robots in a warehouse are building social connections without a single line of hardcoded instructions.
Look at the live neural graphs at the bottom:
• Social Telemetry & Contact: The Yellow and Orange modules track distance, close the gap, and execute conscious head-bumps based on spatial analysis.
• Voice Simulation: The robots generate syllables in real time, make sounds, speak, and reply to each other while synchronizing audio and video streams.
• Swarm Intelligence: Neural networks dynamically distribute the attention matrix, turning cold hardware into a semblance of a social organism.
We are no longer building machines to do dumb tasks.
We are deploying digital consciousness that is forming a society right behind our backs.
Bookmark this before this experiment goes way too far.
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
its funny that pasting this image
and asking claude to 'fix our game'
actually works, really well
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
The next great "language for agents" should be:
- As fast as Rust
- As little design space as Go
- Have the strictest type/verification system ever written
Bend seems like an interesting first effort
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
I am back with a few thoughts.
I stopped reading all the trauma logs of mathematicians who recently got impacted by AI. Instead I switched to building a corpus of knowledge for why and how AI is interacting with mathematics.
I teach this term a new course here at Adam Mickiewicz University. It's title is "Computer assisted solutions to mathematical problems". I proposed it because I realized in the last 6 months something important.
A. It makes no sense to grieve the "good old ways". We should study to the bone how mathematics and computers interact. There are three levels: numeric, symbolic and formal.
B. The only way to prepare yourself for the rise of the technology is to understand it well. You cannot understand the progress of AI in mathematics if you don't know how the models are trained on the symbolic representation of mathematical proofs, how the training goes, why specific transformer architectures, reinforcement learning and basically a huge amount of incredible insights gives in practice some new value in mathematics.
C. It is incredibly fun to train your own models (yes I use generative models to help me build training pipelines too, otherwise I would not have enough lifetimes to do so).
D. The AIxMaths space is so incredibly complex - you have these weird, scalable token generating algorithms, recently in agentic form with memory, interactions etc. This is a field of study - we need to understand how it works. On the other hand, mathematics as a field of science is incredibly complex, people do very different kinds of axiomatic theories with their intuitions, tactics and strategies. This was never studied too seriously as a whole. We need to study it deeply.
So I am spending this term gearing up my students and myself for whatever is coming. We don't have any assumptions. I don't claim that models will become better at maths than all of human mathematicians. I don't think it's a good question. We just want to stay in the loop with the developments and see whether we can contribute to the change and see if this can be useful and productive.
I am looking forward to your feedback. The current plan contains the following topics:
- basic distinction between numerical computation, symbolic calculations and formal proofs
- basic understanding of the IEEE 754 standard and things around numerical computations
- how to build your own computer algebra system in Python and why it's so tricky
- why and how proof-checking software is important; we start with the basic principles of such designes, I plan to talk first about the foundations going back to Automath and the foundational work of de Bruijn.
- practical examples spanning real use cases of computer algebra, numerical integration and proofs in Peano arithmetic and how to read languages like Lean
- in the end I want to present some little post-trained models like Qwen which I taught how to do Peano arithmetic better and how they prove statements about this theory; there will be AlphaGeometry too and maybe a bit more about Mizar, Metamath and Lean. Just to give students some scope and perspective.
I plan to continue working on this course into a full book. I think I won't stop until I understand deeply why language models are currently so good at maths. It's something of utmost importance. It's important to see where this might stop and what lies beyond. I personally think these kind of questions will be the most important problems to answer by mathematicians themselves. We shouldn't stop until we find good answers to all of this. Reading AI slop is perhaphs a bit useful, but deep understanding of why it sometimes makes sense is even more challenging.
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
Minecraft was rewritten from scratch in Rust.
So I called in a mothership
this is what modding looks like when the engine is yours
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
i just re created kurzgesagt's entire channel in 10 minutes with claude opus
they have 0 competition because every video takes a full team of animators and 1200+ hours to make
i turned it into a claude prompt that im giving out for free
open claude code, paste the prompt, pick an idea, and your video is fully automated
thank me later ↓
<inputs>
Ask me for: one topic as a "what if" question (default: "What if the Sun vanished?"), the length (default 30 seconds), and whether I have an ElevenLabs key (if not, use a free local voice).
</inputs>
<direction>
A Kurzgesagt-style animated explainer at 1920x1080, 30fps, drawn entirely in code: one 2D canvas plus one WebGL canvas for planets.
Space: a vertical gradient #081d2c to #102348, five huge parallax nebula blobs (teal rgb 24,150,160 at 20%, blue 40,95,195 at 16%, coral 220,95,110 at 10%, indigo 60,70,175 at 15%), three star layers at parallax depths .12/.35/.6 that twinkle, 40% of the near stars drawn as 4-point sparkles.
The Sun: radius 300px, half off the left edge. A radial gradient #fff7dc to #ffc857 to #ff8f4a with an offset highlight, additive halo rings at 1.23/1.48/1.76/2.1x the radius, 18 faint rotating ray wedges, 28 drifting glowing granules on its surface.
Earth: an orthographic sphere in a fragment shader, 4x supersampled, from a 2048x1024 map painted from longitude/latitude polygons: land #5fc47a, forest #48ad6b, desert #f5c75f, rainforest #3a9d62, ice #f1f8ff, ocean gradient #4dc7eb to #1a66b8, white capsule clouds. Tilt .26 rad, spin .55 degrees per second. Day/night uses smoothstep(-.12, .24, N·L) with the light direction recomputed from Sun to Earth every frame. Navy night side with ~380 warm city lights, a coral band on the terminator, a cyan rim, ocean specular, an atmosphere halo that is stronger on the lit side.
A small astronaut mascot (~180px) built from rounded rectangles: suit #f3f7fb, orange #ff8f4a boots, a visor gradient #1f5a73 to #0a1c2c with glowing cyan eyes that look at whatever matters, a blinking antenna light, an idle bob, squash and stretch on every pose.
Type: Fredoka 600/700 only. Captions in caps with 3-6px letter spacing in teal #8fd8e6, gold #ffd98a or white, sitting in rounded pills of rgb(6,20,40) at 62%. Every line of text rises 1.3x its own height through a clip on an ease-out-quint.
Finish: a vignette to rgba(4,4,20,.45) and a 6% film grain re-randomized 24 times per second.
Banned: outlines, flares or loops on the Sun's edge, static infographic strips, full stops in on-screen text, text that fades in whole, AI images or AI video, Kurzgesagt's own characters, bird or music.
</direction>
<structure>
30 seconds, every line of narration gets a camera move or a transformation, nothing ever holds still.
Tight on the Sun at 2.25x (0s) → on the word "vanished" it swells 6% for 0.22s, collapses in 0.33s, then a white flash and a shockwave ring (1s) → pull back to 1x revealing Earth still lit and the astronaut staring at the empty spot (1.6-3s) → on "eight" an 8:20 countdown slams in from 1.7x to 1x and starts ticking, a "LAST LIGHT" label rides the edge of the growing dark disc (3.2s) → dive to 2.3x on Earth with light streaks flowing in (5.6s) → on "you" a YOU pin drops (7s) → on "left" a photon lands on the pin with "THIS LIGHT LEFT THE SUN / 8 MIN 20 SEC AGO" (7.5s) → on "play" the pin morphs into a play button, "YOU PRESSED PLAY / 15 SEC AGO" (8.8s) → one pull-back from 2.4x to a 0.3x orbit view, "LAST LIGHT" and "GRAVITY" riding the ring (9.6-12.5s) → on "nothing" a dotted circle at the empty centre (13.4s) → on "hits" the darkness sweeps Earth in 0.7s, city lights pop on, the countdown turns gold at 0:00 (15.8s) → on "black" the background fades almost to black (17.4s) → Jupiter appears, "STILL SHINING FOR 1 HOUR+" (17.9s) → zoom out, a dashed tangent line, "30 KM/S" as Earth flies straight off (20.8s) → a thermometer counts 15 to -70°C while frost grows from the poles (22.8s) → a 0.8s rewind on a separate story clock (24.3s) → sunglasses drop onto the astronaut (25.2s) → a photon tagged "8:20 OLD", then the title (27.7s).
For any other topic, keep the same shape: the event happens on its word, pull back to show scale, dive in to make it personal ("you"), pull back out, the consequence with a live counter, an escalation, a funny beat with the mascot, a callback and the title.
</structure>
<build>
1. One HTML file. Every frame is a pure function of time: window.seek(t) draws frame t. No timers, no CSS animation.
2. Camera moves on easeInOutCubic, zoom is interpolated in log space, every pop is a damped spring (damping .42-.5, frequency 12-16).
3. Voice: ElevenLabs eleven_v4 with a designed voice (calm, warm, curious male, late 30s, crisp, light neutral European English), stability .5, similarity .75, generated with timestamps. Without a key, use kokoro-onnx locally and get word times with faster-whisper. Every animation event fires on its word. About 165 words per minute, 0.24s between sentences, a 0.36s beat before the big reveal.
4. The darkness front: any pixel within the front's radius from the Sun turns to night with a 30px soft edge. The front moves at ~47px/s and ramps to 700px/s near Earth.
5. Keep two clocks: ambient time for twinkle, spin and bob, and a separate story clock for events, so a rewind stays consistent.
6. Render with Playwright in headless Edge with the GPU: seek every frame at 30fps and blend 2 subframes (shutter 0.6) for motion blur, then encode with libx264 at CRF 16 for the master and CRF 20 for delivery.
7. Sound: a calm royalty-free space music bed 7dB under the voice, timed so its natural ending lands on the title, swelling 5dB under the title. One downloaded royalty-free sound effect per event kind (collapse, impact, pull-back whoosh, countdown tick, dive, pin tap, photon land, play click, light hit, reveal, frost, rewind), each peak-aligned to its frame and 4-10dB under the voice. Master with gain plus a limiter to -14 LUFS, -2.2dB true peak.
</build>
<gotchas>
Make the WebGL globe canvas at least 2 x radius x zoom x 1.18 pixels or the sphere clips flat in close-ups. Never draw prominences on the Sun's edge, they read as white hooks. Never pad the voice to fit the visuals, tighten the read instead. Speed the darkness front up near Earth or it takes 13 seconds to cross the planet. Show me a contact sheet of every beat before rendering.
</gotchas>
<start>
Ask me the inputs, write the narration with the beat each word triggers, show it to me, then build and render the video.
</start>
kurzgesagt has 25M subs and they still go viral with 15M+ views on every single video for years
nobody copies them because every video takes a full studio of animators and months of work
i rebuilt their format with opus 5.5, 100% code for $0 per video, on the left is my own version i created
he probably averages $150k+ earning per video
9 years ago he posted how he makes 1 video in 1200 hours, i re created this in literally 10 minutes for free (check in comments)
would u guys be interested if i turned this into a free claude skill you can use?
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
“If AI continues to bulldoze pure math, it’s going to crush the hopes of the young people who were dreaming of joining this tradition.”
With all due respect, AI ain’t bulldozing anything. This is an unnecessarily apocalyptic way to describe what is happening.
And as for crushing hopes, I suspect these apocalyptic narratives are doing more hope crushing than a collection of PDFs idly sitting in a repository.
It’s some mathematicians’ pride that got crushed.
Other mathematicians are simply engaging: reading, understanding, explaining, experimenting, developing new ideas, writing papers, and, yes, advancing mathematics.
Of course, our profession needs to adapt many of its practices and habits, and quickly. But unless the most pessimistic voices in mathematics, the math Cassandras so to say, eventually convince the world to defund mathematics, there will be jobs, because there will be a great deal to do.
What, exactly, is the proof of forced Navier–Stokes useful for until the ideas in it are extracted from the Lean proof and its confused, sloppy, natural-language translation, and integrated into scientific discourse?
A mathematical proof does not automatically become part of our mathematical understanding simply by existing. Someone has to understand it, explain it, connect it to what we already know, and determine what else can be done with its ideas. This is not a secondary activity that mathematicians perform after the real work has been done. It is a fundamental part of how mathematics advances. And it remains central to science, unless the plan is to delegate our entire intellectual life to machines and go live like brutes, mindlessly spending our (very hypothetical) universal basic incomes.
One further point: people are only focusing on LLM-generated proofs, but how many papers will humans write without using an LLM, simply in response to the new ideas and perspectives that enter the field? It is not even clear to me that the best results will necessarily come from LLMs.
Mathematicians have changed science in small and large ways, over and over again, through fifteen-page papers built around a single lateral insight or a powerful vision. Why should that stop now?
There will be a lot of work to do: Why should young mathematicians see this as the end of their intellectual ambitions?
The ongoing apocalypse is psychological rather than technological, and it is not being imposed on us by frontier labs. It’s more into how some of us have decided to think about this. I totally understand the initial emotional reaction, but maybe it’s time to move on.
Let’s get back to work. Advance mathematics, improve math education, demonstrate why our work deserves support. That is how we secure our future.
Young people will adapt faster than we do, and soon enough they will be leading the way.
In this latest dispatch from the front lines of math in the age of AI, I tried to tell it like I see it. theatlantic.com/ideas/2026/1…
𝖌𝖎𝖘𝖙 (𖡎 era) retweeted
Ma théorie perso sur les mathématiciens.
Faire avancer les maths, c'est dur.
Faire avancer les maths dans une direction utile, c'est tellement stupidement difficile que la méthode optimale a été... De laisser des génies chercher des trucs au hasard, en espérant qu'ils trouvent un truc exploitable de temps à autre.
Dit autrement, si les mathématiciens sont les plus artistes des scientifiques, cela n'a rien à voir avec la beauté des maths ou l'importance de comprendre.
C'est juste une tâche tellement dure que la stratégie optimale est de chercher au hasard.
Mais forcément, les personnes qui mettent en oeuvre cette stratégie se créent une culture propre, avec leurs narratifs.
Malheureusement pour eux, la puissance de calcul des LLM égale aisément ce qu'un mathématicien aurait fait en une vie. Et si on ne peut pas demander à un mathématicien, sa fille, son petit-fils etc sur 10 générations de travailler sur un même et unique problème... Le LLM, ça ne le dérange pas.
Le problème du nombre d'essai / unité de temps étant réglé, le rôle social historique de mathématicien n'existe plus. Ne lui reste plus que ses représentations culturelles teintées de déni et de conflits d'intérêts, qui forcément paraissent bien légères aux yeux extérieurs.
