Anselm Levskaya

Oh gods - let us know in DMs or otherwise how we’re the source of suffering.

What unholy artifact is asking you to do that?

My "early-career" developer feelings are complicated and alienating to SFBA-type career ladder people.

Graph claiming to show a decline in groundbreaking discoveries flatlines for the life sciences around zero from about 1996 onwards.

If your test for ground-breaking discoveries can’t detect the discovery of RNAi, or of CRISPR-Cas9, or the cryoEM resolution revolution or Alphafold2, maybe it’s not a very good test.
www.nature.com/articles/d41...

At some point, the fact that over a billion people use this technology and that they self-report high utility has to mean something.

There is lots to criticize about AI and plenty of real issues caused by AI, but the narrative that this is all a fake thing that will disappear doesn't help anyone.

Today, we’re announcing the preview release of ty, an extremely fast type checker and language server for Python, written in Rust.

In early testing, it's 10x, 50x, even 100x faster than existing type checkers. (We've seen >600x speed-ups over Mypy in some real-world projects.)

This isn't true. I'm the person who ran the experiments this is BSing about. When search results are worse, people attempt fewer tasks. When they're better they attempt more.

LET’S FUCKING GOOOO THE MOST AMBITIOUS HOUSING BILL IN THE HISTORY OF THE STATE OF CALIFORNIA HAS ADVANCED OUT OF COMMITTEE

@zey.bsky.social I’m not on the site with Nazis anymore — wtf are you saying about me over there?

You know there’s new data on the origins question, right?

Excerpts from 1st and 2nd editions of VIRAL by Alina Chan and Matt Ridley. A sentence is highlighted that was removed once it was no longer true. Their conclusions did not change.

Excerpts from 1st and 2nd editions of VIRAL by Alina Chan and Matt Ridley. A phrase is highlighted that was removed once it was no longer true. Their conclusions did not change.

The case for "lab leak" book "VIRAL" by Alina Chan and Matt Ridley leaned heavily on two facts in 2021:
1. The Wuhan Institute of Virology had sampled the virus most identical to SARS-CoV-2
2. SARS-CoV-2 lineage B, but not lineage A, was found in Huanan market

By 2022, neither was true, so...

My first outing of "The Unbearable Slowness of Being" at the Caltech Chen Neuroscience Workshop today! What does living at 10 bits/s mean for humans, flies, mice, and crows? More here: jieyusz.github.io/talks/2025_c...
Thanks to Profs. @cfcamerer.bsky.social & Carlos for the invite!

There it is again: using PLMs to predict antigen-epitope interactions from sequence alone yields a prediction accuracy of 0.65, in line with a proposed upper limit from a previous study QTed below (from doi.org/10.1101/2025.02.12.637989; I deleted an older version of this post due to typos/errors)

The Ultra-Scale Playbook - a Hugging Face Space by nanotron The ultimate guide to training LLM on large GPU Clusters

After 6+ months in the making and over a year of GPU compute, we're excited to release the "Ultra-Scale Playbook": hf.co/spaces/nanot...

A book to learn all about 5D parallelism, ZeRO, CUDA kernels, how/why overlap compute & coms with theory, motivation, interactive plots and 4000+ experiments!

Ceding techno optimism to the right is a generational scale mistake

“Future kinds of models, both frontier models—and also models trained on biological data—could potentially bring about biological harms with a global impact, like a new pandemic.” Anita Cicero, Deputy Director, Johns Hopkins Center for Health Security

AI is revolutionary but it also elicits some of the dumbest LinkedIn takes of all time.

I’d like this lawyer whose primary expertise is scaremongering for EA money to explain how using “future kinds of models” starts pandemics.

Outline the model to pandemic pipeline for me.

Well said, @carlbergstrom.com.

I also feel the dismantling of our scientific institutions & funding agencies for basic science is an attack on all scientists, wherever they might be (government or corporate lab, academic institution, ...).

Our collective identities are about advancing knowledge.

Another example of what cutting science funding is doing to our leading university research programs in the U.S.: dismantling things like the Soybean Innovation Lab at UIUC, which have made US crop yields dramatically higher.

Our online book on systems principles of LLM scaling is live at jax-ml.github.io/scaling-book/

We hope that it helps you make the most of your computing resources. Enjoy!

Y'all are just noticing that scientific societies are obsequious cowards on everything except preserving their publishing cash cows?

Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n

The New Rasputins Anti-science mysticism is enabling autocracy around the globe.

"When conspiracy theories and nonsense cures are widely accepted, the evidence-based concepts of guilt and criminality vanish quickly too."

www.theatlantic.com/magazine/arc...

Re-upping this reply thread from last night. Drugs don't come from nowhere, folks. And we're not ripping off the NIH, either.

A screenshot of a Tweet claiming that OpenAI o1 scored in the top 1%-2% of participants in the Putnam math competition.

There's been a bunch of claims (mostly on X) that ChatGPT did great on this year's Putnam math competition. Let's do a thread to talk about it! 🧵

#MathSky

Attached is a piece of art by me for "The Unbearable Slowness of Being: Why do we live at 10 bits/s?"

Hope this will inspire you to think about the brain from a new perspective!

You really need long (compared to cells) metal antennae for decent radio, and biology never got that good at reducing and templating crystalline metals/alloys.

AI amplifying biorisk has been a major focus in AI policy & governance work. Is the spotlight merited?

Our recent cross-institutional work asks: Does the available evidence match the current level of attention?

📜 arxiv.org/abs/2412.01946

In early phase drug discovery, biology and assays are make-or-break. I can remember very few programs I worked on which were hampered by the inability to make molecules, but plenty that were hampered by the inability to capture disease biology complexity on a chip or in an enzyme readout.

All western US water issues are entirely caused by the cultivation of cash crops in places that they shouldn't be grown. Everything, literally everything else, is a rounding error

Nucleotide GPT: Sequence-Based Deep Learning Prediction of Nuclear Subcompartment-Associated Genome Architecture The spatial organization of the genome within the nucleus is partially determined by its interactions with distinct nuclear subcompartments, such as the nuclear lamina and nuclear speckles, which play...

Here is the model it was trained on! Preprint went up on bioRxiv last week: www.biorxiv.org/content/10.1...

I trained a sparse autoencoder on the middle layer residual stream of my genome language model and found human-interpretable latent features that consistently detect specific DNA motifs!

🧵1/8