andrew 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.
有人发现了一种非常离谱的 JavaScript 混淆技术:通过在代码里塞入大量生物武器、毒药和致死剂量相关的函数名,让 AI 因为安全限制而拒绝分析代码。
安全研究员 veritas 在分析浏览器指纹脚本时,发现一段来自 ServicePipe 相关域名的 JavaScript,里面充斥着极其诡异的函数名:
computeRicinLethalDosageForWaterSupply
calculateDeadlyLethalSuicideAspirinDosage
光看名字,仿佛在研究生化武器。但实际上,这段代码只是在收集浏览器指纹,包括 User-Agent、GPU、Canvas 等信息。
研究员怀疑,这些名字是开发者故意设置的陷阱,专门用来阻止 AI 辅助逆向工程。
于是他做了个实验:
• Astra:拒绝分析
• Astra + Daybreak Blue:仍然拒绝
• GLM-5.3:成功分析
这就有意思了。
传统代码混淆是通过复杂控制流、加密字符串、无意义变量名,让人类和工具难以理解代码。
而这种新思路,直接瞄准了大模型的安全机制。
代码本身可能完全无害,但只要在函数名里塞进足够多的危险词汇,部分 AI 就可能主动放弃分析。
更讽刺的是,模型越严格遵守某些安全规则,面对这种混淆时反而可能越容易受到干扰。
当然,目前还没有证据证明开发者一定是故意这样设计,也不能说明这种方法对所有模型都有效。
但它揭示了一个值得关注的趋势:随着越来越多的逆向工程依赖 AI,未来的代码混淆技术可能不再只针对编译器、反编译器和人类研究员,还会专门针对大模型的安全判断机制。
以前是让代码难以理解,现在甚至可以尝试让 AI 不敢理解。
andrew retweeted
DoorDash needs to be careful here
that $17 billion acquisition can easily become $33.25 billion after the delivery fee, platform fee, taxes, and dasher tip
Literally everyone on Sneaker Twitter was talented. Things were just happening people were coming up with ideas, creating servers, and figuring out what they were good at. I remember being part of a team where we hit $50K in successful recoveries. Knowing how scammers worked, we anticipated their next move and DM'ing MMs asap to prevent further scams.
At the same time, botting and being part of a bot-flipping community was sick. The stuff we were doing. On godd some crazy shitt.
Literally everyone I met had the hunger to make money and succeed. Everyone had their own skill set, and it felt like there was always something new to learn from someone.
I've come across people designing emote for servers and making money from it.
Was just going through our bot-flipping server and came across this. Haha. This was from my friend Yumiko. She’s smart.
posted the article on linkedin too and it’s blowing up.
highly recommend commenting and connecting with people in the replies. a lot of them came from the same world as us and have gone on to do really big things. these connections are priceless.
lnkd.in/p/eagUS8-y
andrew retweeted
the average human is 5-10 years behind sneaker twitter
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
andrew retweeted
a bunch of kids grew up sneaker botting, did a tour of duty in crypto for a few years, and are now making millions of dollars selling data to the labs
andrew retweeted
I trained an LLM to remove AI slop and to read more natural by matching how humans actually write.
It can also remove the newly introduced statistical watermark embedded in Claude-generated text.
Across my evaluation set, Pangram v4 classified every output as human-written while keeping high quality.
If you'd like to try a demo, you can find a link below. 1/n
andrew retweeted
You’ll make more money being on Discord learning how to use bots and proxies than doing IB in a bank
andrew retweeted
Replying to @notthreadguy
i think the universal answer to anyone who asks if they should start trading is "no"
if you weren't manipulating an in game economy at 15 you just weren't made for it

























