the hardware lottery is also the hardware curse.
with our team, we've been pushing the boundaries of what fully local neural cellular automata models achieve on visual reasoning.
we believe this opens the door to a swathe of efficient, robust, and scalable hardware designs.
Can complex multi-step reasoning emerge purely from cells that only talk to immediate neighbors?
Happy to share our paper “Reasoning with Neural Cellular Automata (NCA)”, from our team at Google, Paradigms of Intelligence 🧵👇
Alive Maps exists in NYC. it's called @moseymaps (mosey.world/app).
it plans the walk you'd actually want (scenic, shaded or sunny) and you set how many extra minutes it's allowed to detour.
reply with two NYC spots and I'll send back a walk worth the extra 8 minutes ↓
today I discovered a walking route to work that’s 2 minutes longer than the maps default, but takes me past a park, beautiful architecture, and a street full of cafés and life. such a huge upgrade to a journey I make every morning
I want Alive Maps, a version of Google Maps that optimises for beauty, serendipity, and aliveness instead of speed
Research News! Announcing the 1st release of fundamental AI research from the Open-Endedness Team @LilaSciences: a vision-focused blog + 4 preprints. The blog details a broad research agenda for achieving open-endedness while each paper unveils its own novel research direction.
The ideas covered in the papers include:
• a novel approach to training for creativity designed to sidestep a newly-identified problem called the “creative paradox,” which means that directly reinforcing creative outputs can sometimes limit reaching full creative potential;
• an investigation of the advantages of seeing the world through fresh eyes by training an LLM that is intentionally naïve in a certain area (in effect presenting less data as a potential advantage);
• a new kind of training regularization technique for SGD called unified factored organizers (UFOs), which is designed to achieve unified factored representation (UFR) as seen in Picbreeder CPPNs;
• and finally a unique study of catastrophic forgetting that shows that networks with UFR (like those from Picbreeder) are intrinsically resistant to catastrophic forgetting. The inset video on this post is a visualization from that catastrophic forgetting work.
The blog post also includes an interactive “wheel of science” demo.
The implications of these ideas are broad and diverse, but the fun part is that each one opens up its own new research playground - there is a lot of room for further research here!
Moreover, we hope this release also highlights one other important point about AI today: while it may seem that the road ahead is already paved and all that’s left is to drive mindlessly forward, there actually remains a lot of room for new directions and insights in the pursuit of open-endedness and creativity.
This research was possible because @LilaSciences appreciates the critical role of open-ended innovation in the pursuit of scientific superintelligence. The advances you see here are part of this much larger company-wide ambition, which includes AI Science Factories and many other unprecedented components. When this engine is fueled by open-endedness, the result will be much more than a step change.
More advances in open-endedness are coming!
Blog link below 👇
My AI agents found an unmapped Maya site in Mexico, using LiDAR data from a NASA dataset.
72 hours from a random thought on vacation to a reply from Mexico's national archaeology institute.
What a time to be alive…
Transformers being like "I have no idea what this is about, but I minimize cross-entropy better because I leverage every single statistical cue I spot in my flawless recall of the conversation".
check out Kunal's latest paper! understanding swarms, self-organisation, and its advantages in agentic settings is a super exciting new (...old, but renewed) frontier!
What if a swarm of AI agents could learn to work together on its own? 🐝
No leader. No roles. No turn order. Just one shared reward.
Our new paper s³: multi-agent RL lets LLM swarms self-organize, beat best-of-N, and scale past their training size!
scalingswarms.com/ 🧵
1/4) We're hiring in my lab, **please share**! 🚀
There are two postdoc opportunities in cog neuro and AI.
This is for a new project supported by a Schmidt Polymath Award. This project aims to build an engaging video game to collect data for neurofoundation models. 🧠🤖
1/5
Continuous diffusion may be better at discrete reasoning than standard sampling suggests.
We took SRM's MNIST-Sudoku Hard checkpoint (arXiv 2502.21075) and changed ONLY the sampler:
❗ 0.8% → 65.7% valid boards solved, without retraining. Why does standard sampling fail?🧵
Proud to share new work from our Paradigms of Intelligence team toward AI designed for decentralized hardware, where low-power communication and fault tolerance are critical requirements. Using Neural Cellular Automata (NCA), @mayalen_etc and her co-authors show that spatial locality is not a barrier to complex multi-step reasoning.
Read the full paper on arXiv: arxiv.org/abs/2609.36126v1
Can complex multi-step reasoning emerge purely from cells that only talk to immediate neighbors?
Happy to share our paper “Reasoning with Neural Cellular Automata (NCA)”, from our team at Google, Paradigms of Intelligence 🧵👇
Biology has been doing computer science for four billion years. Before the discovery of DNA, Alan Turing and John von Neumann worked out the logic of self-replicating machines, showing how matter folds into a system that reads and executes instructions to build itself.
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
enabling these microbots to "walk" was largely a materials engineering breakthrough btw
quite clever to use platinum's electrochemistry to induce surface stress, and then cap with titanium to make the whole film bend
we're teaching AI to do this kind of science @periodiclabs
in case you missed it, nanobots are here and they can be fabricated on standard semiconductor equipment
one silicon wafer can produce millions of micron-scale robots that can move, think, and communicate
turns out when you have too many GPUs for your own good you can do stupid stuff such as pre-training a foundation model with constrained decoding on PTX -- that is, it does not speak natural language, it's tokenizer is PTX, it's thoughts are PTX, it's all PTX 🧵
This week marks one year since “What Is Intelligence?” was published.
It was a really fun book to write. I knew that some of the material would be dated by the time it reached readers. But I hoped that, despite rapid progress in AI, the core ideas about life and intelligence would endure. So far, so good, I think!
Hearing how you have connected with the book has meant so much. Thank you for reading, and an extra big thank you to those of you who have reviewed it, recommended it, or given it to friends.
Can an LM, starting from random init (!!), learn to generate all of its pretraining data?
Introducing Self-Play Pretraining with Zero Data. Two models start from random initialization: a generator proposes programs for a universal Turing machine and a learner trains on their outputs. We never train on any real data, but see predictable scaling on natural datasets: zero-shot val loss on images, text, audio, and melodies decreases predictably with self-play compute. And the learner develops in-context learning capabilities.
A fun proof-of-concept, co-led with @AdityaCowsik and @KfirDolev and co-authors @gbruno_dl, @ANourya@noahdgoodman, and @YoavLevine.
1/ New essay with @blaiseaguera and James Manyika at the DeepMind Institute: AGI may not arrive as one titanic mind but as societies of agents. The hard problem shifts from building a mind to orchestrating, governing, and living inside many of them. institute.deepmind.com/essay…