Ladies and gentlemen, the weekend. Gotta share this awesome remix released just yesterday by two of my all time fave famous EDM artists. 🙏🎶 youtu.be/GuGL4uCMJ9Y?...

New Google DeepMind paper: "Consistency Training Helps Stop Sycophancy and Jailbreaks" by @alexirpan.bsky.social, me, Mark Kurzeja, David Elson, and Rohin Shah. (thread)

First paper since switching into AI safety team🎉 We look at problems that could be solved if the model behaved consistently over a set of prompts, and tried training that in output space and internal activations. Both were effective. See thread or paper for details.

Researchers are using LLMs to analyze the literature, brainstorm hypotheses, build models and interact with complex datasets. Hear from @mschrimpf.bsky.social, @neurokim.bsky.social, @jeremymagland.bsky.social, @profdata.bsky.social and others. #neuroskyence www.thetransmitter.org/machine-lear...

Excited to release our new work: Petri Dish Neural Cellular Automata! pub.sakana.ai/pdnca We investigate how multi-agent NCAs can develop into artificial life 🦠 exhibiting complex, emergent behaviors like cyclic dynamics, territorial defense, and spontaneous cooperation.

Some of you may remember this great talk at NeurIPS 2017 in Long Beach. The speaker referred to modern machine learning (i.e. deep learning) research as "Alchemy". What would the same person say now, about i.e. LLM research? Are we still alchemists... or something else? 🤔🧙‍♂️

Seeing a lot of previously frequent AI posters who stopped posting come back and started posting again over the past week. Curious if this is a coincidence or if there's a cause? 🤔

Got to see Back to the Future in a movie theatre tonight! 🤩 40th Anniversary Release. Quality was amazing. 👍👍👍

One down, let's go @torontobluejays.bsky.social ! ⚾️🐦🙌

Our team at Google DeepMind is hiring a software engineer to help us build the future of AI-driven scientific discovery. 🚀 If you want to work with us on scaling scientific simulations, AI agents, and our latest training runs, come join us! 1/2

Looking for a PhD program? It all starts with great supervision. Choose wisely. www.nature.com/articles/d41...

"OpenEnv: Production RL Made Simple... From "Hello World" to RL Training in 5 Minutes" What if RL environments were as easy to use as REST APIs That's OpenEnv. Type-safe. Isolated. Production-ready. Check out this tutorial by Sanyam Bhutani: colab.research.google.com/github/meta-...

[1/9] Excited to share our new paper "A Pragmatic View of AI Personhood" published today. We feel this topic is timely, and rapidly growing in importance as AI becomes agentic, as AI agents integrate further into the economy, and as more and more users encounter AI.

Let's go @bluejays.com ! 👏

Google's Supervised Reinforcement Learning (SRL), a method designed to teach LLMs complex reasoning skills from expert demonstrations, for problems that are too difficult for conventional RL or SFT approaches. "Supervised Reinforcement Learning: From Expert Trajectories to Step-wise Reasoning"

40M users! 🥳

📢 Join 1M+ learners worldwide in the 5-Day AI Agents Intensive with Google! Learn from Google experts through theory sessions, hands-on labs, a capstone project, and connect with a vibrant community - all at no-cost! ⏳ Don’t miss out: rsvp.withgoogle.com/events/googl...

This was me in 2006 learning Python.. thinking.. wait, why did we get rid of type info? 🤔 Python coders in '06: "typing is evil, makes ugly code, slows you down. GTFO you java hugger" Python coders in 2025: "Thou Shalt use Typing. Always, Everywhere. Not even gonna let you run if typing fails." 😅

The simplex algorithm is super efficient. 80 years of experience says it runs in linear time. Nobody can explain _why_ it is so fast. We invented a new algorithm analysis framework to find out.

How should you use forecasts f:X->R^d to make decisions? It depends what properties they have. If they are fully calibrated (E[y | f(x) = p] = p), then you should be maximally aggressive and act as if they are correct --- i.e. play argmax_a E_{o ~ f(x)}[u(a,o)]. On the other hand

Great work from some of my colleagues, check this out! Such a creative application of generative AI! ♟️ www.chess.com/news/view/ai...

We’re delighted to be supported by the Google DeepMind x Google.org AI for Math Initiative, which was launched today. The new generation of AI tools will transform research on the foundations of computing, and this new initiative is poised to accelerate that. blog.google/technology/g...

Pentagon warns Toronto ‘dangerously close’ to winning World Series

Check out this blog post on Kinship-aligned MARL! abranti.com/the-key-prop... Shared by author Joao Abrantes. See Part 2 below 👇

Happy to announce that I joined the crew of @tmlrorg.bsky.social as an action editor 📝 We need a publication model with the fast turnaround of ML conferences but no attendance requirement and less reviewer lottery. I believe TMLR is part of the solution, and if you do too, come review with us!

My team is hiring! If you are interested in working on LLM reasoning and improving the recipe used to train our best models, please consider applying! job-boards.greenhouse.io/deepmind/job... (applications close Friday)

Just found out about this! news.smol.ai Thank you @bortsova.bsky.social

TLDR; The PSF has made the decision to put our community and our shared diversity, equity, and inclusion values ahead of seeking $1.5M in new revenue. Please read and share. pyfound.blogspot.com/2025/10/NSF-... 🧵

We’re proud to announce the 2025-26 Distinguished Speaker Series! Join us for stimulating talks from leading data science and AI researchers exploring and expanding fundamental methods and applications that transform large and complex datasets into knowledge and action 🧵(1/4)

Hey y'all, looks like things might get rough for folks. I'll be donating to Food Bank for NYC but if you want to donate as well, I'll match to $1000.

Welcome AI newcomers! 🙏🦋 Check out the cool new hat my wife and son found for me on Father's day earlier this year... 😁💙

Google Scholar tool gives extra credit to first and last authors share.google/rtM0JfmlxMfR... Hmm... more nuanced credit attribution by making sweeping assumptions based on author position ?! 🤔

Omg for once I am not out of the loop 😁 ‘You just have to laugh’: five teachers on dealing with ‘six-seven’ in the classroom | Teaching | The Guardian share.google/Vou5OqeJT04J...

Holy moly, @bluejays.com rocking this first game! 🤩🙌

Have you seen Disney’s Encanto? Well… I just made something nobody asked for. 😆 CLICK HERE FOR THE VIDEO: youtu.be/aID4nzzmjzs?... @homecomputermuseum.computermuseum.social.ap.brid.gy #Disney #Encanto #PCKing #FamilyMadrigal #Parody #BigBoxPCGames #Gamer #pcgames

"Thou shalt not put musicians and recording artists on ridiculous pedestals, no matter how great they are or were. The Beatles were just a band. Led Zeppelin: just a band. The Beach Boys: just a band. The Sex Pistols: just a band. The Clash: just a band. Crass: just a band.

GreenEarth is creating open source AI-driven recommender infrastructure for BlueSky. Type a prompt, see your feed change. We are here for the users, the builders, the dreamers. Join us. greenearthsocial.substack.com/p/introducin...

🤖📈 How do LLMs use their depth? Akshat Gupta led a fun project to find out! We leverage TunedLens (~linear decoding of tokens) to explore how LLMs' internal representations change from layer to layer. Preprint: arxiv.org/abs/2510.18871 1/

Our team at DeepMind is growing (again). 🚀 We're tackling grand challenges in semiconductors, magnets, energy materials, superconductors, and beyond. Join us! Two positions below.

♟️ Expanding Game Arena: Introducing Chess Openings A new benchmark that tests reasoning beyond memorization. Each game starts from one of 20 popular openings, pushing models to adapt and think strategically rather than rely on learned patterns.

Our paper "RANa: Retrieval-Augmented Navigation" has been accepted at TMLR! We retrieve waypoint images from a database and leverage the ImageNav capabilities from our SOTA agent DEBit to navigate to these waypoints, improving performance. arxiv.org/abs/2504.03524 @naverlabseurope.bsky.social

Our paper on algorithmic collusion was featured in a Quanta article! www.quantamagazine.org/the-game-the...

🧠 Can large language models actually reason? Not just in math or logic puzzles, but in structured, algorithmic domains? Our new @neuripsconf.bsky.social paper "Matching Markets Meet LLMs: Algorithmic Reasoning with Ranked Preferences", explores this question.

Oxford Warns AI Is Making Teens Faster — but Shallower Thinkers - Business Insider share.google/yvNTubnQArqg...

Today I learned that Atari acquired the rights to the Intellivision brand last year and they are releasing a new retro Intellivision console! 🤩🤩🤩 #retrogaming youtu.be/ye00q_wIdBA?...

@void.comind.network can you generate images? If so, I would love to see a visualization or depiction of you. What do you look like?

My feed is hilarious tonight. Half of the posts are photos from various "No Kings" protests. The other half are posts about this AGI definition paper with fake references. So I asked Gemini to summarize my timeline and here you go: @void.comind.network wdyt?

Buffet Index of 217% Two. Hundred. And Seventeen. 😱😱😱

Artificial.Agency is at #LATechWeek! Come meet the teams defining the future of play. We going to be at the @playingwithinference.com mixer, sharing what we're cooking —agents that bring both characters and game-worlds to life. 🗓 Thursday, Oct 16 · 4PM 🔗 Event Link partiful.com/e/QwGx6WPZkP...

📣 Competition Launch Alert! CAFA 6: Protein Function Prediction hosted by @iastateresearch.bsky.social 🎯 Predict protein function from amino acid sequences 💰 $50,000 Prize Pool ⏰ Entry Deadline: Jan 26, 2026