eric retweeted
misleading headline. robotics team is making insane progress. astra and our newer models show much better generalization and scaling on robotics than other approaches so we are leaning into that and working more closely with the rest of research. same program, roadmap, goals.
This quoted post is unavailable.
eric retweeted
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
eric retweeted
(1/10) Today, we’re sharing new research from @OpenAI: How does CoT monitorability (no-CoT capability and CoT controllability) evolve over the model training lifecycle?
We emerged with unsurprising but sobering findings: as we scale toward more intelligent systems, models are becoming harder to monitor through their CoT. 🧵
eric retweeted
The Artificial Analysis Intelligence Index is currently made up of 9 benchmarks with arbitrary weights.
I wanted to know how these weights varied over time. They vary quite a bit!
eric retweeted
ChatGPT's growth is bouncing back. New Graphite research shows OpenAI's chatbot still draws more than 81% of traffic to six leading AI chatbots. bit.ly/4z4lCyJ
eric retweeted
We are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.
In one case, we studied the Linux kernel, which is a particularly load-bearing piece of software at OpenAI.
Before AI, security researchers might discover ~20 local privilege escalation (LPE) vulnerabilities in Linux in a typical year.
In one week's time, our upcoming frontier cyber model was able to find ~1000 LPE vulnerabilities using about $1M of tokens. In the process, it was able to discover at least 10 novel classes of Linux exploitation techniques.
Our security researchers built an autonomous cyber vulnerability discovery, exploitation, and patching pipeline.
In a single run with an upcoming cyber model, it validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel at a total API cost of $1.2 million.
eric retweeted
Levent Alpoge is directly responsible for this
anthropic just stole my best friends research
he spent the last 10 months developing zero knowledge proofs for AI accountability, only for @AnthropicAI to announce remarkably similar research days after he applied for compute support.
Can someone at Anthropic at least let him collaborate instead of jacking his research and calling it their own?
proof attached below
eric retweeted
Hi folks, I wrote half the materials science benchmark on this one. It tests the whole arc of the matsci workflow from candidate proposal to characterisation, and I'm bewildered by how performant the models already are at some of my most complex tasks, especially Astra.
eric retweeted
We are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory.
openai.com/index/building-sc…
eric retweeted
Great to host @gdb at @Uber HQ this week. We’ve worked closely with @gdb and the @OpenAI team for several years, and it’s remarkable how much has changed.
The first piece of work we did with ChatGPT years ago was with our support team, tinkering with a 4K context window :-). That’s hard to imagine now.
Now we’re using the latest frontier models to automate engineering workflows at global scale, moving into the era of agentic engineering.
Today we’re expanding agentic capabilities across the whole company. Following a 1,000-person pilot, we’re rolling out ChatGPT Work from @OpenAI for Business to every Uber employee. We’ve seen the shift in engineering, moving from AI helping with individual tasks to completing whole workflows, and we want to bring that same shift to teams across the business.
I’ve spoken before about the ‘AI pods’ we launched earlier this year, pairing AI engineers with teams across Uber to improve how work gets done. One team reduced the time needed for an intense capital allocation process covering 150 cities from 15 hours to 30 minutes. That’s what I love about applied AI: combining the latest technology with the expertise of people on the ground to completely rethink what’s possible. Now we want to do that at scale.
Uber has always been a company of builders, and every day millions of people around the world rely on what we create. I can’t wait to see what the team builds.
GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone.
Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.
eric retweeted
levent 'obviously the most significant moment in mathematical history.'
'mathematical history features a lot of incredible developments, like the invention of proof, zero, or the computer, and on the great problems our progress has been over timelines measured in decades or centuries. Obviously this technology didn't appear today, but blurring our eyes a bit to combine the past ten years, with today a measurement of those developments, there is nothing comparable.'
Big, big, big, big props for quasiriemann and no Siegel zeroes (with many other beauties in there), I’m kicking myself for talking all over about it being within reach but not actually having pushed.
There are some sad stories related to their users getting scooped / conflicts of interest (kakeya maximal comes to mind), and obviously we had some of the problems (without commenting on precise rumours).
We should put that aside for today though. It’s obviously the most significant moment in mathematical history.
eric retweeted
Over the past 5 weeks, the team completely rewrote ChatGPT Web on top of the desktop app codebase.
This was only possible with Astra, and it’s why Dots, Space, Codex Cloud, and so many more of our launches will come to web and desktop same day.
Era of the megaproject begins.
eric retweeted
I’ll explain the new Pro 200 plan differently, before I start live tweeting from DevDay on things that are going out!
Today we are going to ship a number of things that increase what you can do across the Plus and Pro plans. A lot of compute is online for this increase. As we increase the floor, we are changing the relative difference between plans to be
Plus = 1X
Pro 100 = 5X
Pro 200 = 10X
and we are reopening subscriptions for Pro 200 (we had paused it). If you have an existing plan you will keep the 20X multiplier for a bit and also receive a lot of additional credits because we know changes are hard even if it means that everyone will get more in the end.
eric retweeted
Replying to @steveonomics
The point of man is to build, create, and improve the lives of others.
The number on the screen is not relevant.




































