
AI @hyperspell | t̶w̶e̶e̶t̶s̶ posts are my own unless they are bad, in which case I disavow
San Francisco, CA
Joined April 2017
- Tweets167
- Following288
- Followers93
- Likes14
Andrew Burkard 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.
Maybe being forward deployed wouldn’t be so bad. 🤔
Today we’re launching full EU data residency at @hyperspell
Data sovereignty is critical for your company brain. European companies need confidence their data stays local
Storage, AI processing, backups, and logs can stay in the EU
Your data stays in Europe. Where it belongs
1
2
142
Andrew Burkard retweeted
introducing @hyperspell live pages: your company’s self-updating wiki
connect slack, gmail, notion, and granola, and your wiki updates as your team works
one source of truth for your whole team and agents, with every fact linked to its source
we made an emo music video for the launch
Andrew Burkard retweeted
great having the @Pelion_VP crew by the office
excited for what's ahead
Always a treat catching up with the YC Batch Daddy himself @contextconor, who also coined “single until Series B,” and his incredible cofounder @maebert
Two early pioneers of the “company brain” movement building @hyperspell, now building one hell of a team! Cheering you both on!
i’m at @modal runtime. come say hi if you want a single until series b hat.
6
452
One delightfully unexpected side effect of knowing my Slack activity may be slurped up by an AI brain is that it’s made me a slightly more specific communicator. I find myself specifying time zones, replying in threads, and using proper nouns.
still not using proper punctuation, capitalization, or spelling tho. those are irrelevant
70
One thign i've noticed recnetly is that using LLMs has made be a significantly sloppier typer. You canreally jsut mashthe keys ina ny old order and the silicon brain will be able to figure it out.
1
63
My weekend project: using frontier-level intelligence to figure out how many touches the Titans third string running back is going to get this year. 🏈
I used GPT-6 Astra to build a draft model for my fantasy football league, and it kinda, sorta, somewhat beats the expert consensus in some backtests.*
(link in reply)
1
128
Shout out to everyone at @UseCorgi, especially Hamilton. 🐶
Andrew Burkard retweeted
The most advanced use of @usefiretiger I've seen has come from @hyperspell
AI slop emails have more in common with actual Spam than classic junk emails ever did: ultra-processed, mode-collapsed sludge.
Say what you want about boner pills and Nigerian princes, but at least they had range.

















