
Fusion energy computational research engineer/physicist, AI for Science lead and head of digital engineering @PPPLab
Princeton, NJ
Joined March 2013
- Tweets7.3K
- Following2.4K
- Followers853
- Likes18.8K
Congratulations to Levent and Anthropic more aptly put
The Lean FRO and Mathlib Initiative have released a joint statement on the Lean verification of the Riemann Hypothesis.
On November 20, Levent Alpöge released a proof of the Riemann Hypothesis as a Lean library of 17 million lines of code. Over the past day, the Lean and Mathlib communities worked together to verify it. The statement certifies the proof as correct and describes how it meets all necessary verification criteria.
Congratulations to Levent on this achievement, and thank you to everyone in the community who contributed to the verification.
Full statement: leanprover-community.github.…
16
Michael Churchill retweeted
It’s all over but the shouting. In the span of a single day that will echo through mathematics history, the solution to math’s hardest problem was announced—and then independently verified.
sciam.com/article/the-ai-rie…
Michael Churchill retweeted
Flow Matching connects two distributions.
Can a one-step generator learn from the flow between its own samples and real data?
Meet Midpoint Generative Models (MGM), our new training framework built on this idea.
Here’s how. 🧵
"What Levent's proof seems to have done instead is something completely different. It takes an incredibly complicated tour through the infrastructure built for a hundred other mineral reserves, situated far from the destination, and somehow create a wormhole to the original 1/2
Like most mathematicians, I believed the Riemann Hypothesis to be completely out of reach at present (even with the best AI tools), because I expected it to require substantial theory-building, probably in arithmetic geometry. The kind of new theory-building that fills entire books, and that AI isn't so good at yet.
Mathematicians have a notion of how deep a problem is. A sufficiently deep problem is generally intractable until enough of the theory is in place. If difficult problems are like valuable mineral reserves buried miles underground, you first need to build the infrastructure to reach them.
What Levent's proof seems to have done instead is something completely different. It takes an incredibly complicated tour through the infrastructure built for a hundred other mineral reserves, situated far from the destination, and somehow create a wormhole to the original destination.
This is not theory-building in the usual sense. It is something much weirder and more alien. The best human mathematicians know how to build theory, painstakingly, to reach a goal. But no human brain, on its own, is capable of creating a wormhole that requires a 25,000 page long tour through existing infrastructure with no obvious or realistic connection to the goal. But huge credit to Levent here: I think no current AI, on its own, is capable of this either.
I had never worked on or thought too much about the Riemann Hypothesis: it seemed very far removed from my own area of number theory.
But the astonishing thing about Levent's claimed proof of RH (which I think has a very good chance of being correct) is that it uses a huge number of precisely the things that are central to my own corner of number theory:
- Theta functions and other automorphic forms on unitary groups
- Gan–Gross–Prasad period formulas
- Relative trace formulas
Etc., etc.
A problem that I had always assumed would require new theories in arithmetic geometry may instead turn out to be tractable through an inhuman chain of existing (deep) results in higher-rank automorphic forms (and a lot else).
I've thought about this a lot. I never liked when colleagues searched for solutions or thms because they were "beautiful" (often meaning simple or compressed), implying correctness.
Replying to @MilesCranmer
we're discovering all the things we were myopically blind to when we went around saying "any solution to X must be elegant and short" when the universe is far from elegant
674
Michael Churchill retweeted
BREAKING: A mathematician at Anthropic claims to have used the company's AI to solve the Riemann hypothesis, the most notorious open problem in all of math
sciam.com/article/anthropic-…
Michael Churchill retweeted
Today in AI science:
* Generalized Riemann Hypothesis: alpo.ge/remark.pdf
* unforced euler cims.nyu.edu/~tristanb/cc.pd…
* unforced Navier-Stokes blowup github.com/tristanbuckmaster…
Michael Churchill retweeted
We introduce Posterior Matching Distillation (PMD), a new member of distribution-based distillation for few-step generation. PMD trains a stochastic student whose samples reproduce the teacher's exponentially tilted means, each obtained from a single frozen-teacher evaluation at an input shifted along the tilt direction.
We prove that matching these tilted means over a neighborhood uniquely determines the teacher’s full posterior over clean images. The result is exceptional diversity across seeds for the same prompt.
w/ Julius Berner (@julberner), Arash Vahdat (@ArashVahdat), and Pinar Yanardag (@PINguAR).
🧵 1/5
Michael Churchill retweeted
“Real-time” video models are finally real-time!
Introducing PixVid: lifting a latent video model to generate one frame at a time, in pixel space. No VAE, no temporal compression.
The payoff? Fast high-res generation, precise editing, and ultra-low control-to-screen latency!
Would be a huge mistake to hire a CEO without a technical understanding of chip manufacturing and AI ecosystem. Jensen aligned all the hardware companies, said this is where the field is going, gave a direction and backed it up.
Hot take: The people calling for Jensen to be brought back are hurting NVIDIA at this point.
He isn't coming back. I don't believe there are enough shareholders coming to vote to fire the board and extend this shit show.
So where does this lead us?
A world of pain.
As if the last week or two hasn't been enough.
Get ready for more pain.
But at some point capitulation will happen. Everyone will see that bringing back Jensen won't happen.
Then what?
Investors should, instead, be looking at the depth of leadership in this company. Looking at its strengths and asking themselves "what kind of leader do we want to see put in place?" Who can come in and get everyone to see a new vision for NVIDIA?
Is there a leader on the field who can keep NVIDIA's strengths and minimize its weaknesses.
I found one yesterday.
Ferrari's CEO, Benedetto Vigna, who spoke about the role of innovation at the Slush Conference yesterday. Which was quite excellent, by the way.
He knows a lot more about that, and running a company that has a leadership position, than anyone else I can see.
You might knock him about how he launched Ferrari's electric vehicle, which is sold out now, by the way, and paves the path for more innovation at @Ferrari.
Convince him to join as CEO of NVIDIA. That would get the world to see NVIDIA has a leader who can continue its innovation and leadership role.
This "bring Jensen back" movement just isn't gonna lead us to a good place.
As much as I believe he was wrongfully, or, at minimum, awfully, fired, that ship has sailed and it's time to find a new leader who can step in and get NVIDIA headed in a new direction.
This "bring Jensen back" kind of belief just will increase NVIDIA's pain and extend it for years.
Time for new, bold, leadership to take NVIDIA out of the morass it is in.
166
Michael Churchill retweeted
Introducing FieldCraft 🌊
github.com/tangency-labs/fie…
Another free, open-source project from Tangency Labs, spun out of SolidCraft, by @jyviko: one modeler for structures, heat and fluids, in pure Rust. Parametric part in, verified results out.
- Meshes follow the exact CAD geometry, with no tessellation facets on curved parts
- Automatic hex and tet meshing, including O-grids for holes and revolved tubes
- Lattice Boltzmann CFD with curved walls, checked against pipe and Couette flow
- Heat transfer with radiation, including solids in a flow
- Every result checked against theory, for engineers and AI agents alike
Michael Churchill retweeted
Latent diffusion needs no decoder😵
New paper: DWD, a linear head on DiT predicts pixels directly
• Same or better quality on every tested latent model, less flicker in Wan video
• Even pixel models improve: ImageNet FID 1.55 ➡️ 1.43 (pixel SOTA), photorealistic texture in T2I
Excited for this work led by Jacob Merson (RPI), neutronics calculations are critical for fusion device design, and algorithms to leverage GPUs and accelerate detailed, high-fidelity calcs can change how we approach design
sciencedirect.com/science/ar…
131
Michael Churchill retweeted
Having been in the image generation community for this long, I've noticed some tricks that may boost your generation performance at no cost.
All the aggressive opinions below are my own, but credit for the findings goes to my amazing friends and collaborators @XiaoyuW9401147 @Weihang404 @lowerbad.🫡
Michael Churchill retweeted
EmbeddingGemma 2 was the #1 trending model on HuggingFace for over a week. A spot usually only held by LLMs.
What it is, how it works, and how it was trained: [1/6]
Michael Churchill retweeted
Replying to @make_robot_go
I think it’s 3 hours not 3 minutes help but yes SuperMimic is crazy
supermimic.github.io/
Michael Churchill retweeted
Last couple years I've been asked what I think about decoder-only vs encoder-decoder for multimodality, and never really given an answer.
Ellis and Chris take a good step towards answering that question in this study.
Two big points:
(1) Recent encoder-free models like Gemma4-12B, Inkling1, TUNA-2, etc are all a single "point estimates": they show this is possible but tell us absolutely nothing about how it scales, or doesn't! -> this paper does.
(2) Encoder-free or not is actually a false dichotomy. It hides three questions which Chris and Ellis nicely disentangle, and answer separately:
(a) separate pre-training,
(b) separate vision params, and
(c) vision params before or inside decoder?
Answer in pic below, and full thread and paper.
Labs disagree on vision encoders for MLLMs! ⚔️
Gemma 4 12B scrapped the encoder.
Kimi K3 dropped encoder pretraining.
Inkling 2 Small revived the encoder Inkling killed.
We trained 200+ models in a controlled study to settle it:
✅ Keep vision parameters, skip the pretraining. And they don't have to be an encoder! 🤔
Paper: arxiv.org/abs/2611.17334 🧵
Michael Churchill retweeted
We keep the pretrained DiT and replace the VAE on both ends.
Encoder → a pixel patch embedding, initialized from a linear fit to the VAE encoder, with SNR-matched timestep calibration.
Decoder → a lightweight causal conv head that renders DiT tokens into pixels.
Michael Churchill retweeted
And of course 𝐒𝐩𝐚𝐠𝐡𝐞𝐭𝐭𝐢 𝐒𝐩𝐥𝐚𝐭𝐭𝐢𝐧𝐠 can do a lot more than... spaghetti – check this bee out – data courtesy of @DanyBittel
𝐑𝐞𝐚𝐥 𝐭𝐢𝐦𝐞 𝐯𝐢𝐞𝐰𝐞𝐫 for this video can be found here spaghetti-splatting.github.i…
📢📢📢 Introducing Spaghetti Splatting
3DGS is a convolution of a Dirac with a Gaussian Kernel. We provide a strict generalization of 3DGS, extending its Dirac mass to lines and curvilinear structures
project page: spaghetti-splatting.github.i…
arXiv: arxiv.org/abs/2611.12716
🔊 on!





















