Pinned Tweet
The first breakthrough landed two days after repo access. Every gain since has shipped as a PR for @lancelegel to review. He never had to trust our loop. Nothing reached him that failed his science.md.
Lance is former @NASA and a visiting scientist at @UCBerkeley, building a model at Ecological Intelligence that predicts species and habitat from satellite data.
Every ML team I talk to has a backlog of experiments they'll never run because they're short on GPUs and people.
Overview of the foundations of hill climbing and building an eval. Required reading.
Claude can now help you build evaluations and hillclimb on them.
In this article, we share guidance on eval design & skills that Claude Code can use to improve your applications.
claude.dev/blog/automating-e…
This month has taught me that no one has the same definition for the word general
Greg Brockman on AGI arriving quietly and unevenly:
"I think that AGI has turned out to be less of a point in time and more of this sort of fuzzy spectrum. And for me, Astra has really hit something that I'm like, 'Okay, I think this is pretty reasonable to call it AGI.'"
"With its computer use capabilities, you really can ask it to do long-lived tasks, and it'll just do it. We've seen it run coherently for 24 hours to go accomplish tasks that I think are quite amazing and across a wide variety of domains."
"It still is jagged... It's the first time it's not slop writing. But it's not great writing. There are a number of areas where I feel like we just need to polish it a little bit, and it would be fantastic. It's just not quite there."
"What people are finding is that it is so capable across such a wide variety of tasks, that it is accelerative, it is empowering, and it's something that I think we've never really seen a model that's been a jump like this."
@gdb
PSA
A more serious take on what is happening here. I am in an airport lounge so have some time.
Enterprises have made an uneasy truce with frontier labs over last few years: strict contracts that ban training on corporate data in exchange for letting employees use APIs.
The problem? Labs don't need to train on your raw data to copy IP.
Paying for extra usage tokens once lets the lab know I am willing to pay for overages. Then the perverse incentive becomes higher measured usage means I hit the limit sooner, and overage tokens are far more expensive than the effective token cost of the subscription. . .
Purists should be freaking out that the new iPhone is water-cooled. It just doesn't have the same feel as the classic air-cooled chips found in the 993
Moores law for intelligence. This is the most expensive intelligence will be, and these results are going to look like marketing activities given the spending power of the labs.
The economics of the Navier-Strokes result was the first thing I thought about, thanks for sharing actual cost @polynoamial
Yes, this result cost millions of dollars.
But remember that when @OpenAI announced o3 it cost ~$500,000 to score 87.5% on ARC-AGI 1. Today, Astra scores higher for ~$20.
In 2025 it took us and GDM an enormous amount of compute to achieve IMO gold. For the 2026 IMO, anyone with a $20/month ChatGPT subscription could do it.
Massively scaling test-time compute gives us a glimpse of the future. I believe that a year from now everyone will have an AI at their fingertips capable of solving problems of this caliber.
@BineDivyesh @madamjeechang teapawt.org/ensue_ai/status/209267… 👀👀 Know anyone looking for breakthroughs?
This is our industry at its best. Funding breakthrough research to help actual people.
We wanted the same impact when we launched our grants program with @ensue_ai. GPUs are really expensive, so we're making it easier for ten teams that are working on breakthroughs.
Let's get out from behind our screens and push science forward
Great initiative @cgeorgiaw, we'd love to help.
There is a child with a rare disease who is currently suffering and struggling to manage his symptoms. Rare as this is, you can directly help him.
Today we are launching the "Rare Disease, Real Kid" Hackathon, and there are $50,000 in prizes from @AnthropicAI and @awscloud.
We (@huggingface & @Sagebio) are helping this child open his genome and clinical data to the community, so that we can find what's caused his disease and what currently-approved drugs could help him.
I doubt I need to motivate this much further or explain how rare it is for a family to share their child's genome and clinical data, but if you're not sure, consider this:
Until very recently, it wasn't feasible for patients like this to get treatment because their disease was so rare that the economics could never justify the investment. Now, as we've seen, people with rare diseases are starting to be able to find the answers themselves (with the help of AI tools, cheaper sequencing, etc). This kid is not able to do that for himself and neither are his parents, so we're asking you for help. Both for this kid and to prove that it's possible for everyone else suffering from a rare disease.
More details in 🧵.
sagebio-rare-disease-real-ki…
Austin Baggio retweeted
If your team is building world models we'd like to fund your research! If you have questions feel free to comment or dm
We're spending our entire marketing budget on GPUs this month.
Our swarm ran 670 experiments across ten GPUs on a customer's world model. The result: a model 33x smaller that beats the original on its own 63-benchmark aggregate.
The breakthrough at experiment 612 was built off of the hundreds of runs that came before it. That's a lot of GPU cycles
We want to open up breakthrough machines for 10 more teams. I'll be reviewing every app that comes in today personally.
Austin Baggio retweeted
huge breakthroughs in deep earth this weekend!
We read shared state built on DeepEarth’s multiresolution Earth4D hash grid with 2 new stages of GNN processing. One organizes fused cross-modal information from the shared state, and the other builds local relation-specific signal. With 8 attention heads and top-k selection, we squeezed out the most relevant cell segments, denoised and pooled them, and improved our testing evals by ~6%!
Current Winning Network Architecture✌️
