The advent of LLMs has subtly tried to convince us over and over that struggling is not worth it, but in reality, the most learning happens through the struggle..

Seems like something’s fundamentally wrong with my #bouldering skills as I sway back and forth a lot 😂 Meanwhile I’ve climbers with smooth well thought out steps

You can get MSAs directly from AlphaFold DB now (alphafold.ebi.ac.uk). I also missed the FoldSeek implementation, which lets you search for structurally similar proteins direktly. This saves some clicking around. Neat!

Tobias Kreiman, Yutong Bai, Fadi Atieh, Elizabeth Weaver, Eric Qu, Aditi S. Krishnapriyan: Transformers Discover Molecular Structure Without Graph Priors https://arxiv.org/abs/2510.02259 https://arxiv.org/pdf/2510.02259 https://arxiv.org/html/2510.02259

I don't think I'll ever cease to be amazed by the fact that The Global Internet is not that ethereal, abstract, thing, but instead hinges upon 500 very concrete big fiber optic cables running on the ocean floor.

❤️ nice piano progression here

A serious and useful thread with permission to take AGI seriously by @jascha.sohldickstein.com over at X x.com/jaschasd/sta...

view from Eagle rock trailhead was breathtaking ❤️

My takeaway: Their Approach == ESM2 (3B) + SimpleTransformer w. Flow Matching (X B) When X=100M, it falls way short of ESMFold performance (10-15%) When X=3B, it closes the gap ---> Still, AF2+MSA (95M) is far superior But it's nice to know how far simple transformer+flow matching can take us

having a fun time TA'ing.. despite the fact that it takes time out of my research, it gives a nice way to regain energy in a different way... I want to co-teach a course one day before I graduate 😆

RFdiffusion3 is here! We train a general network that explicitly models every atom and use it to design active enzymes and DNA binders. Tremendous team effort with Jasper, Rohith, Raktim, Rafi, Yanjing, Paul, Jonathan, and many others! Check it out: lnkd.in/eiUFfJaM.

Started practicing a Korean song 😆 reading lyrics is harder than I thought genius.com/Genius-roman...

my toxic trait is wanting to hear the gpus sound while working -_-

would be nice to have something like usegalaxy.org, for structural bio tasks.. GUI to connect RFDiffusion output with LigandMPNN etc..

v1.108 is rolling out today 🚚 Now live, at long last: Bookmarks, aka Saved Posts. For all those posts you'll definitely plan to come back to! Update the app and give it a try. The button is right down there 👇

if you want to see how determined someone is.. ask them to compile dgl library from scratch for aarch64 😅

aarch64 is so hard to deal with :/

two things to master well 🤔 1/ what questions to ask 2/ where to look at

All I want is an AI agent that handles my slurm jobs across clusters

Being kind boosts mental health more than seeking joy. Evidence: Doing 3 random acts of kindness a week is enough to reduce depression, anxiety & loneliness. It's more beneficial than doing nice things for yourself. Self-care feels good, but generosity builds lasting bonds.

I'm excited to share FlowMol3! The 3rd (and final) version of our flow matching model for 3D de novo, small-molecule generation. FlowMol3 achieves state of the art performance over a broad range of evaluations while having ≈10x fewer parameters than comparable models.

This is Spud. She was offered a stick of celery today and took personal offense to that. 12/10 (IG: spud.the.cryptid)

Good to be back in Pittsburgh ❤️

"Pat, why do you carry that ridiculous 600mm lens on long hikes?" Buddy, I can see mountains reflected in the eyes of a trailside pika.

time to time, it's good to take some time off and just reflect on those who had a big impact on your life, and send a quick note updating on what you've been up to

RFdiffusion2 is now live! github.com/RosettaCommo... You can now design proteins, and in particular enzymes from just partially defined amino acid side chains, and without defining their sequence position or order!

FlowMol3: Flow Matching for 3D De Novo Small-Molecule Generation [new] 3D De Novo Small-Molecule Generation: Flow matching model improves molecule generation using self-conditioning, fake atoms, & geometry distortion.

mol-view-spec’s Jupyter and streamlit integrations are kinda cool github.com/molstar/mol-...

RosettaFold 3 is here! 🧬🚀 AtomWorks (the foundational data pipeline powering it) is perhaps the really most exciting part of this release! Congratulations @simonmathis.bsky.social and team!!! ❤️ bioRxiv preprint: www.biorxiv.org/content/10.1...

all i want is a jupyter environment that goes with me where every i go, from one server to another

a quick guide to copy data from one remote to another with rsync: gist.github.com/ramithuh/f4d...

Job alert 🚨 Our team is looking for ML Research Scientist to join Johnson&Johnson research. We work on geometric deep learning and LLMs in drug discovery. 🧬🤓 Drop me a message if you’re interested, or share this if you know someone who’s a great fit! Multiple locations available.

Asking the right questions isn’t easy

🌳 Nothing says summer at NEC Labs America like our annual Princeton picnic! This year’s gathering brought together colleagues for a day of connection, great conversation, and plenty of fun. 🍧A big thank-you 🙏🙏🙏 to everyone who came out and made it such a memorable afternoon!

This is Haha. He had to socialize multiple times today. Gonna need a couple days to recover. 14/10 (IG: thegoldenhaha)

an advice that can never go wrong "leave a place better than you found it. No matter where you go."

The drums on Wildflower (Billie Eilish) is so nice 🤯

thanks to jet lag, have been waking up at 5.40 AM consistently for a week 😆 then spend 6.00 -6.30 AM --> thinking about the day gotto admit 8AM - 11AM may be the most peaceful time to do deep work at the workplace 🧘‍♂️

Diffusion LLMs are promising ways to overcome the limitations of autoregressive LLMs. Less error propagation, easier to control, and faster to sample! But how do Diffusion LLMs actually work? 🤔 In this video, let's explore some ideas on this fascinating topic! youtu.be/8BTOoc0yDVA

Excited to share work with Zhidian Zhang, @milot.bsky.social, @martinsteinegger.bsky.social, and @sokrypton.org biorxiv.org/content/10.1... TLDR: We introduce MSA Pairformer, a 111M parameter protein language model that challenges the scaling paradigm in self-supervised protein language modeling🧵

internet in Sri Lanka is a test of patience 🫠 (at least in my home) on a remote machine, even the terminal completions are slow :|

managed to do a temporary fix while the machine is shut down... 😃 while I cannot index new posts, that doesn't mean we cannot serve the old ones.. So pushed all the posts indexed up to now (7465) to a static GitHub repo... while my machine is moved, a static snapshot will be served

Can AI actually make us worse learners if we use it too soon? We should earn the privilege to use automation—basics first. A real story of AI in education gone wrong, and why process matters. Seen AI help or hurt learning? Share your thoughts! ericmjl.github.io/blog/2025/7/... #aiineducation

This feed will have a 𝗱𝗼𝘄𝗻𝘁𝗶𝗺𝗲 on Jul 23, 2025 (Wed). Sorry for the inconvenience! (The machine is moving to a new room 😅, maybe it's time to use a cloud machine 🤔) If you have any comments, suggestions on the feed, visit the GitHub repo! github.com/ramithuh/mls...

4xH100 > (8xA100s ≈ 16xTitanV)

Figured out multi node training for the first time 🥵 Cluster network configs gave me a hard time though…

slurm should have an algorithm to prioritize users who has had historically higher gpu_utlization percentage for their jobs 🤨

in the future, in the AI+product space, there will be a demand for tools that encourage the user to think.. soon we will all realize that tools such as Claude Code, Gemini-CLI are making developers' thinking somewhat inferior [if not used the right way]