I am an AC for ICLR 2026. One of the papers in my batch was just withdrawn. The authors wrote a brief response, explaining why the reviewers failed at their job. I agree with most of their comments. The authors gave up. They are fed up. Just like many of us. I understand. We

🚨New package alert! 🧬 Introducing Pocketeer: A lightweight, fast protein pocket finder in pure Python! I built this as a Pythonic version of fpocket because I wanted something that could easily calculate subpockets. > pip install pocketeer Link at end of the thread 🧵(1/7)

This is the century of biology. We can create synthetic neural networks inside of living cells. We can turn skin cells into egg cells and use them to create babies. We can generate drugs by prompting a digital superintelligence that runs on the cloud. We can edit our genomes

Cameo'd with some goats 🐐 It was a super fun evening ... so many spicy takes :D

Beyond structure-based biomolecule design: Dynamics, black-box data, and the antedisciplinary frontier of biomolecule design Isomorphic Labs is founded, AlphaFold 3 has a very different release strategy than previous iterations, and there now exist multiple biotech (techbio?)

this take is pure BS and misses how deep tech innovation actually works Ilya has a PhD in CS from Toronto under Geoff Hinton where he co-invented AlexNet & literally helped birth the modern DL revolution before founding OpenAI Adam has degrees in CS and Mathematics & built

Productivity culture exists to kill the exact behavior that leads to breakthroughs. The people who build important things look unproductive because they waste hours on random shit nobody asked them to solve. Strategic planning is cope for people who can't tolerate obsessive play

Imagine talking to your WandB dashboard before going to bed to intervene and re-run an expt with diff hparams if something bad happens overnight. Fascinating ...

Beyond Structure-based Biomolecule Design 🧬🚀 Its an important moment: models starting to work, multiple biotechs building similar protein design foundation models, and action shifting from academia to industry. My new substack article: chaitjo.substack.com/p/beyond-struc…

Structure-based antibody design... in sequence space! We trained a guided discrete diffusion model to generate antibodies and predict the shape of their CDR loops Given one experimentally-validated binder from animal immunization, we produced dozens of high-affinity binders,

Today we’re announcing $106M in new funding led by Altimeter Capital and Bezos Expeditions. This brings our total to $150M to scale our frontier AI models which make biology programmable. Our frontier models have generated functional proteins (Nature Biotech, 2023), created the

Today we’re thrilled to announce JAM-2 — the first AI model capable of generating drug-quality antibodies straight from the computer, with industry-leading success rates. > Drug-like affinities: Picomolar to single-digit nanomolar antibody binders for half of 26 targets while

Today, we're launching GPU-accelerated DFT for all Rowan users. Using GPU4PySCF, we see massive speedups for routine DFT tasks, making it significantly cheaper and faster to run high-accuracy quantum chemistry through Rowan. Here's two representative examples:

Today, we’re releasing new data showing that Chai-2 can design antibodies against challenging targets with atomic precision. >86% of our designs possess industry-standard drug-quality properties without any optimization. Thread👇

Fantastic turnout for the BoltzGen seminar in London at Recursion's office last night! Huge thanks to everyone who joined, the energy in the room was great, and it’s inspiring to see many already testing new BoltzGen designs on real problems.

Chai Discovery is out with what looks like very similar results to Nabla Bio's JAM-2 that I reported out yesterday Protein design space making traction on de novo drug-like antibodies, beyond just getting hits/binders on targets Chai's blog: chaidiscovery.com/news/chai-2-mab

Is your LM secretly an SAE? Most circuit-finding interpretability methods use learned features rather than raw activations, based on the belief that neurons do not cleanly decompose computation. In our new work, we show MLP neurons actually do support sparse, faithful circuits!

Chai strikes back 🤨