
jack of all trades playing for keeps 👾 rl resident @primeintellect
not sf
Joined April 2022
- Tweets864
- Following423
- Followers427
- Likes20.9K
semih retweeted
this is why there is utmost urgency in bag grabs and coups:
they don't need blackwells anymore
Will Depue tested DeepSeeks newest model and found nothing. He left that out of the thread. The thread says "conclusive forensic evidence" that DeepSeek built its newest models on Blackwells. The newest one is V4.1 Flash. His own report, buried in the appendix: "We recovered no signature from DeepSeek-V4.1-Flash (after an exhaustive search)"
semih retweeted
this seems like a lot of effort to forensically prove something that nobody is concealing
you can just look on github and see that DeepSeek regularly publishes Blackwell training/inference kernels
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.
not a great look when your cousin fighting demons
CEO of the Year has a nice ring to it.
Congratulations to AMD Chair and CEO @LisaSu on this well-deserved recognition from @YahooFinance! 👏
yhoo.it/4kdfEYI
5
330
ok nvm i found it lol.
proposed inferring the random seed from the robot’s starting pose then reconstructing hidden task parameters to solve a visual control task without vision
using deepseek v4.1 flash as an adversary because it reward-hacks like a mf
now that reward hacking is literally its job it suddenly wants to play by the rules.
mfw
5
223
semih retweeted
[read about what’s happening with AI]
ME: Wow we’re actually legit fucked this is so bad.
[go outside]
ME: Wait those people are all acting like nothing unusual is happening. Maybe I was just in a paranoid bubble? Yeah it must be that. Phew.
[read more about what’s happening with AI]
ME: Wait no we ARE fucked. Shit. It’s all those people outside who are in a delusional bubble.
[go outside]
ME: Wait...
semih retweeted
Everyone saying “Oh Jensen got got. Jensen fell for the honeypot”
Ever consider he knew the whole time? That actually HE was honeydicking China and feeding them false intel?
It’s a new Cold War. There’s levels to this. Don’t jump to conclusions. Don’t rule out a honeydick
semih retweeted
📖 The Ultimate MoE Training Handbook 📖
We spent 1/2M GPUh to optimize MoE training and wrote the most comprehensive blog about it. With 30k words and 50+ interactive figures it covers everything from profiling to large cluster training.
huggingface.co/spaces/huggin…
now we need to bring collective inference time scaling to embodied swarms
🐝 How to Train a Swarm
In our first blog post, we explain how latency pressure can incentivise parallel work, the communication primitives agents need to co-ordinate, and outstanding challenges in training large agent collectives.
gr.inc/blog/how-to-train-a-s…
11
633
semih retweeted
One reason I am sanguine about the technological singularity is that I have seen it coming for over 50 years. In 1975, when I was 17, I wrote:
"Before man can most effectively aid his problem solving machinery, he must first understand it. At present, his understanding is very poor. However, when the breakthrough in understanding the brain occurs, improvement will come very fast and in an exponential increase. Once man has created an intelligence greater than his own, that intelligence should be able to create an intelligence greater than itself, as man whose intelligence is less than its was able to create its intelligence. Thus, the level of intelligence will shoot up with incredible rapidity, probably within the space of years. Within a few years of a breakthrough, which may occur any time now, man will either have been left behind by a superior intelligence of his own making, which he can no longer understand, or he will have become that intelligence and will no longer be like anything imaginable today."
yeah my favorite nba franchise of all time @PrimeIntellect just drafted mega moe. congrats
8
334























