Pinned Tweet
read this, then notice what everyone is talking about instead.
the US bans nvidia's best chips from china. this report shows six top chinese models were built on them anyway & the proof is sitting in the public weight files. method is right on all 24 known cases. not a leak, not a source, just math on files the labs published themselves.
same week nvidia's ceo is gone. headlines say affairs.
my bet: the affairs are the story nvidia can survive. half a billion dollars of banned chips inside china is the one it can't.
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.
Yo is this @clavicular approved?
49
why i think the affairs are the cover, not the cause.
the board cut jensen loose in under 24 hrs. boards do not move that fast over an affair, they move that fast when lawyers tell them the company's exposure is bigger than the ceo's. & nvidia's singapore problem isn't new. for two years the filings showed singapore as a massive share of billed revenue while almost nothing actually shipped there. analysts asked. nvidia said customers invoice through singapore. nobody pushed, cuz pushing meant tanking a huge chunk of the s&p.
in feb a us official said deepseek had blackwells. commerce did nothing public for nine months.
to be clear, none of this proves jensen personally directed a single shipment. nobody has shown that.
but the likely truth isn't "one guy got honeypotted." it's that nvidia, its resellers & the govt all saw the signs & chose not to look, bc looking was expensive. the affair handed everyone a clean exit. fire the man, keep the business model.
the ppl acting shocked this week had the data the whole time.
1
1
55
@grok how strong is the evidence in Will Depue’s report that these Chinese models were initialized on Blackwell chips, and does it support the claim that Jensen Huang’s resignation over the affairs was mainly a cover for export issues?
1
34
I'm someone who has to see a thing to understand it. Reading about it doesn't do much for me. So I built a skill you can use with Claude to take a question and get back a 3D model you can open up and click around in.
.
.
.
Here are four, on four very different things.
4. Something you can't see. This is how GPS finds you while you drive. Each satellite's signal gives one distance, about 21,000 km. Draw each distance as a ring on the ground, and the four rings cross at one spot. That spot is the car.
There are more, and you can open them all in a browser. It's a free skill. I use it with Claude, and it works with other assistants that load skills.
git clone github.com/auswallace/mech <skills folder>/mech
Then type /mech and a question.
14
You know when someone explains how something works, and you nod, and you still have no idea?
I built a fix for that.
1
138
Wallace retweeted
On her final day as DNI, Tulsi Gabbard dropped the Fauci Files.
Proving Fauci:
- Funded risky gain-of-function research in Wuhan
- Lied to Congress about it
- Actively buried the lab-leak truth
CNN’s coverage?
Zero. Not one single mention.
This is the biggest cover-up in modern history, and the corporate media is still protecting it.
Is it because CNN was a knowing participant, or because Big Pharma accounts for ~25% of CNNs ad revenue?





















