Mark Neumann
@MarkNeumannnn
Head of ML at Orbital Materials. BC: Research/Eng at @allenai_org
Claude Code, but there's a small chance that functions you write are antisemitic
Grok CLI >>> Cursor and Claude Code We wanted an IDE for @xai's @grok so we did something meta: we prompted the newly released Grok 4 to create... Grok CLI itself! Grok CLI can: 1. Modify local files and use the shell 2. Go through huge codebases and fix them 3. Persist for…
Announcing Ambient Protein Diffusion, a state-of-the-art 17M-params generative model for protein structures. Diversity improves by 91% and designability by 26% over previous 200M SOTA model for long proteins. The trick? Treat low pLDDT AlphaFold predictions as low-quality data
i was asked by a few (inc. @YuanqingWang ) what i meant by this earlier tweet, and since i'm pretty busy, i decided to write a blog post to answer the question 🤦
ai for drug discovery looks a lot like machine translation research/development during the cold war.
I compiled and replied to responses on my recent blogpost. Several people basically disagreed with all my intuitions - so this is a discussion worth having! @MarkNeumannnn @erikjbekkers @TimothyDuignan @A_Aspuru_Guzik
After a long hiatus, I've started blogging again! My first post was a difficult one to write, because I don't want to keep repeating what's already in papers. I tried to give some nuanced and (hopefully) fresh takes on equivariance and geometry in molecular modelling.
(1/n) Sampling from the Boltzmann density better than Molecular Dynamics (MD)? It is possible with PITA 🫓 Progressive Inference Time Annealing! A spotlight @genbio_workshop of @icmlconf 2025! PITA learns from "hot," easy-to-explore molecular states 🔥 and then cleverly "cools"…
Matbench Discovery is out in Nature Machine Intelligence @ Paper: rdcu.be/esRFB Leaderboard: …tbench-discovery.materialsproject.org No better time to thank all my co-authors @RhysGoodall, @PhilippBenner2, Yuan Chiang @cyrusyc_tw, @Bowen_D_, Mark Asta, Gerbrand Ceder @cedergroup,
🤹 New blog post! I write about our recent work on using hierarchical trees to enable sparse attention over irregular data (point clouds, meshes) - Erwin Transformer. blog: maxxxzdn.github.io/blog/erwin/ paper: arxiv.org/abs/2502.17019 Compressed version in the thread below:
The ML community generates so much incredible software, like I can literally write almost arbitrary functions which are made differentiable by a compiler. And yet we still choose YAML to configure everything (this, Hydra). We have not yet ascended
Put together some notes on modelyaml.org - an interesting attempt by @lmstudio to define a standard way of describing how to run a model - where to get the weights, what architecture and options to use: simonwillison.net/2025/Jun/21/mo…
Announcing Ambient Diffusion Omni — a framework that uses synthetic, low-quality, and out-of-distribution data to improve diffusion models. State-of-the-art ImageNet performance. A strong text-to-image results in just 2 days on 8 GPUs. Filtering ❌ Clever data use ✅
Very cool paper (almost directly refuting a blog post I recently wrote about people not training conservative diffusion models). Would love to know if this scales, would be super cool if it does
What is the probability of an image? What do the highest and lowest probability images look like? Do natural images lie on a low-dimensional manifold? In a new preprint with @ZKadkhodaie @EeroSimoncelli, we develop a novel energy-based model in order to answer these questions: 🧵
In the future anthropologists will study how Dishoom invaded the American cultural psyche, absolutely fascinating - literally every other American visiting London will tell you they want to go Dishoom is fine, but please visit Tayyabs or Lahore Kebab House instead/ as well
I could live here