Read the Stand with Jensen letter and consider adding your name. #standwithjensen #bringjensenback standwithjensen.com/
근데 어떤 모델도 묻지 않는 문제점에 대해서는 이야기 하지 않는다. 사람은 문제를 해결하다가 "이거 너무 복잡해 보이는데...단순한 방법이 있을텐데" 라는 질문을 하기 마련. 근데 모델은 "이거 복잡하지 않아?" 라고 물어야 분석하기 시작.
Fable 5.2는 메타인지에 탁월하단 느낌. "지금 우리가 하는 질문이 맞아?" 라는 식의 생각을 하며 문제에 대한 질문을 던지고 결국 원하는 걸 얻어내는 것. Opus에서 잘 안되던 부분.
점점 이야기가 복잡해진다 🫡 중국에 수출 금지된 Nvidia 칩으로 중국모델들이 train됐다는 포렌식 증거
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.
이번주 구글이 decision model + hydrogen 발표한건 꽤나 실용적인 결과. 프론티어에선 밀리지만, 코딩 말고는 이정도 수준의 agent로도 꽤 많은 어플리케이션 커버할테니.
오늘은 gemini hydrogen + decision model 사용한 agent loop. Decision model (JEV등) 은 프롬프트보고 yes/no 단순 답을 빠르고 싸게 함. 그리고 reasoning model 은 low, high 같은 생각하는 깊이를 설정할수 있슴. 그래서 질문 보고 이게 어려울지 decision에게 묻고 모델의 깊이 설정.
에이전트 가성비로 sonnet 5.5 가 default 였는데 구글에서 드디어 경쟁할만한 모델 나온듯. 지능은 비슷, 속도 빠르고, 가격은 25% 싸게.
Gemini 4 Hydrogen excels at knowledge work.
Based on over 6,000 real-world agentic sessions, Hydrogen (High) is #5 in Agent Arena overall. On knowledge-work tasks, it moves up to #4.
As a result, it’s ahead of three rival models that outrank it overall: Claude Opus 5.5 (Max), GPT-6 Astra (Max) and Claude Sonnet 5.5 (Max).
This means that @GoogleDeepMind's Hydrogen is not only the fastest model in the Agent Arena top 5, but also a powerful choice for real-world work.
Thinkpad는 두가지 정도 아쉬운 점.
1. 끈적끈적 키보드. 맥북의 톡톡 간단한 키보드에서 넘어오면 키감이 좋은걸 넘어서 약간 노동으로 느껴지는 부분.
2. 역시 M계열 cpu 아니고서는 팬소음을 피할수가 없다. 빌드같은 작업은 팬소음 역시. 팬소리 사라지면 AI 가 또 농땡이 피는구나 알기엔 좋다.
박상민 / Sang-Min Park retweeted
A sad day for NVIDIA, developers, America, and all humanity. Jensen founded and ran one of the few companies to reach a massive scale while driving massive continued innovation and ubiquity, while others fell into the trap of ever-increasing extraction.














