Profile avatar
joao.omg.lol
Research Engineer at Google DeepMind A special snowflake existing in 196883 dimensions #ActuallyAutistic He/Him
343 posts 5,918 followers 2,179 following
Prolific Poster
Conversation Starter

If you want a good life, inheriting is becoming nearly as important as working. That is dangerous for capitalism and society econ.st/3QDUfYu

Join Roman and Elliott one last time as they reflect on their journey with you all through The Power Broker, exploring their favorite moments and answering listener questions in this bonus episode.

"The seductive futility of network visualization" skewed.de/lab/posts/ha...

💯% worth a watch if you're into artificial life, complexity science.

Also - government is not a business. It is ultimately a collective that provides services to citizens. It does not need to be run like a for profit entity.

The headline that 21 DOGE staffers have resigned is a little misleading. These were employees of the United States Digital Service, which existed pre-administration and was absorbed into Musk’s team. These aren’t the young idiots; these are actual civil servants.

Excited about our progress in characterizing The Computational Advantage of Depth in Learning with Neural Networks. Check out the number of samples that can be saved when GD runs on a multi-layer rather than on a two-layer neural network. arxiv.org/pdf/2502.13961

Our advance in statistical physics analysis of simple graph convolutional neural networks arxiv.org/pdf/2402.03818 will appear at the Learning on Graphs conference: logconference.org. My student Odilon has now extended this to the multi-layer case, so more is coming soon.

👋 from the Swift team, now on Bluesky!

don’t know who needs to hear this but empathy doesn’t require eagerness. you don’t have to be excited about helping other people, you do it because it’s right and because the alternatives are worse

“The Soul of Craftsmanship” (2025) bencornia.com/blog/the-sou...

Computational-Statistical Tradeoffs at the Next-Token Prediction Barrier arxiv.org/abs/2502.12465 New paper (another fun internship project!) with Dhruv Rohatgi, Adam Block, Audrey Huang (ahahaudrey.bsky.social), and Akshay Krishnamurthy (akshaykr.bsky.social). 1/11

Part 2 of the collaboration with Terence Tao on the cosmic distance ladder is now out. It covers how we first learned the distances to planets, stars, and galaxies far, far away. youtu.be/hFMaT9oRbs4

This is an exciting bit of progress in the governance of OCaml. Historically, too many PRs have languished because there wasn't a clear process to get to consensus. I'm hopeful this will help! discuss.ocaml.org/t/ocaml-lang...

Coming around to the view that a certain kind of narcissist simply needs Musk to be a genius rather than a fraud, because they want to believe there’s a hereditary Brahmin class of accomplished, beautiful minds—and that they’re a part of that class.

#mathematics #math

Do you want to compile OCaml to Wasm? We have you covered: tarides.com/blog/2025-02... This includes support for effect handlers: ocsigen.org/js_of_ocaml/...

Good to see WSJ catching up, not sure why it couldn't have just posted this as its original story on Wednesday though www.wsj.com/science/phys...

We are hiring on the Generative Media team in London: boards.greenhouse.io/deepmind/job... We work on Imagen, Veo, Lyria and all that good stuff. Come work with us! If you're interested, apply before Feb 28.

(Boosts encouraged!) Hi! 👋 I’m Kat. I’ve been a professional dev for over 15 years. I maintained the NPM CLI for 5 of those. I sat in TC39. I work at Microsoft. I’m self taught w/a film degree and no formal CS edu. Are you new to tech? What do you wish you could ask some1 like me? Literally AMA

Obsidian is now free for work. Starting today, the Obsidian Commercial license is optional. Anyone can use Obsidian for work, for free. Explore the organizations that support Obsidian on our site. obsidian.md/blog/free-fo...

THE CROWDFUND FOR NEW EDGE SWORD & SORCERY 2025 IS LIVE Three issues - a Sword & Planet special - a New Jirel of Joiry story... The more re-posts this gets, the more bonus stories will be added to this year's issues! Go here to learn more and back the magazine: www.backerkit.com/c/projects/b...

This Kickstarter for the UK's first fantasy-only bookshop looks super cute! www.kickstarter.com/projects/pap...

This is older than some children who can do math, but still a favorite! I made an animated magic circle to symbolize the Tracy-Widom distribution’s universality — #SciArt for an article by my friend Natalie, for Quanta mag: www.olenashmahalo.com/project/trac... 🧪 🧮 ⚛️ #gif #gifArt #mathArt

PODCAST: Chase was joined by @cado.bsky.social from @remapradio.bsky.social to answer a very important question: What the heck is Android: Netrunner?

There is no space history without Black history. There is no science without Black women. Despite whatever this administration attempts to erase. #BlackHistoryMonth

Hey folks, want a good spooky Point & Click Adventure? Hob's Barrow is one I'd highly recommend.

UK bands from the 80s all have bios like: "The group met at a government-funded arts mixer where they each received six months' rent tax-free" while for bands from the 2020s they're like "formed as a final year project at Goldsmiths, their first release was funded by interest from their trust funds"

I'd point to something deeper I think this thread is touching on: the (US) popular imagination has lost the concept that a system or society can be good, worth representing or defending, the underdog narrative that individuals are good and systems are bad or corrupt runs very deep in US storytelling

Very pleased to share our recent work, in which we use LLMs for automated discovery of interpretable models of animal behavior 🪰🐀🕵️‍♀️ that take the form of Python programs 🐍 See below for a summary of key results by @pcastr.bsky.social!

Can LLMs be used to discover interpretable models of human and animal behavior?🤔 Turns out: yes! Thrilled to share our latest preprint where we used FunSearch to automatically discover symbolic cognitive models of behavior. 1/12

This sounds very cool! The insight appears to be that if you can fool any computationally bounded adversaries, then you can fool any reasonable algo looking at the result of your computations. I.e., see the world as an "adversary", use cryptographic primitives to fool it. arxiv.org/abs/2502.130...

no yeah that's my bad for wanting to open a PostScript file. every CS paper from like 1999 is lost media

RFK Jr. thinks that a clinical trial should be done for the entire childhood vaccine schedule. Here’s why that will never happen. 1. You can’t do a randomized trial of something that we already know is safe and effective. There would have to be a placebo group that didn’t get vaxxed. Unethical.

I love this thread about the impact the National Science Foundation has had on the life & scientific path of @mammalssuck.bsky.social. I too was helped by NSF as an undergrad, through their Undergraduate Research Opportunities Program(UROP). Sad to see what is happening at NSF the past few weeks 😥

Elements of Clojure is now available as a free PDF: elementsofclojure.com