Permabear1993

Skeptical Empiricist. PhD in Chemical and Biological Engineering. 10+ years in biotech. Armchair philosopher and trader on the side.

[Permabear1993 @permabear1993]

@theanantpai Most frustrating is that tech grabs all the attention, and it distracts from actual areas where research funding is needed. Good example is assay development generally, which is chronically ignored (pharma doesn’t fund it, not exciting enough for NSF, not mechanistic for NIH)

[Permabear1993 @permabear1993]

Chinese companies can use compute OUTSIDE OF CHINA to train their models. Whether smuggling is occurring is largely irrelevant.

You can’t stop [short of an insane sanctions campaign] Chinese firms from accessing GPUs.

[will depue @willdepue]

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 https://t.co/YOtzND0Jqk.

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.

[Permabear1993 @permabear1993]

@Satyaki_R You mostly have it backwards,
every capitalist economy has massive regulatory capture and is ran by an elite corporate lobby

[Ron C @RonC345]

People really think better reasoning will solve disease. It’s like asking the smartest person on earth to guess how many fingers you’re holding behind your back.

[Permabear1993 @permabear1993]

@_ClimateCraze @McClellanOsc Being against renewables is proudly declaring ignorance. Also, it’s not a “got em” to accuse someone [who wants to reduce fossil fuel use] of being dependent on fossil fuels - society is dependent on an unsustainable energy system, everyone is part of that society, so no shit

[Ben Norton @BenjaminNorton]

US intelligence agencies claim the Chinese businesswoman with whom Nvidia CEO Jensen Huang had an affair, Alice Huang, is a spy. But there's no evidence of this.

In this video, I explain the real political and economic interests behind the scandal:
https://t.co/YsJtF6u7Jn

This is part of the larger "chip war", in which US hawks and some Big Tech companies want to prevent China from getting access to advanced chips needed to train AI models, in an attempt to monopolize the technology.

Alice Huang was likely smuggling Nvidia chips to Chinese companies like Alibaba, in violation of US export restrictions. Nvidia simply wanted to continue selling to the massive Chinese market.

However, she was probably NOT working for the Chinese government, which has in fact taken actions to restrict Nvidia. Beijing has devoted tons of resources to developing a domestic chip ecosystem, and has pressured Chinese companies to buy chips from local firms, not Nvidia, in order to boost domestic capabilities.

Alice Huang's activities actually undermined Beijing's technological self-sufficiency goals. So why would it recruit her as a spy?

[Ben Norton @BenjaminNorton]

Did Nvidia's CEO Jensen Huang truly have an affair with a "Chinese spy"?

Or are China hawks (and some US Big Tech companies that fear Chinese competition) manipulating this scandal?

I analyzed the evidence, exposing the conflicting interests:
https://t.co/3CU5HBPVUg

[Permabear1993 @permabear1993]

@RebeccaYChan Except that most Americans don’t care where you are from and that’s the truth

[Permabear1993 @permabear1993]

@jonathanchait Wow haha are you a dumb cunt or a disengenuous cunt

[Permabear1993 @permabear1993]

@SGRodriques Wow what nifty plagiarism you have created

[Permabear1993 @permabear1993]

@krystalball claps for all the cynical democrats who smell blood in the water and plan to pretend to be populists until they knife the left like they always do

[Permabear1993 @permabear1993]

Pressley is a cynical opportunist but at least she has balls. Jeffries is awful

[Ayanna Pressley @AyannaPressley]

Do you agree? We need to show that this agenda has strong support. Sign our petition at https://t.co/uD1q3pOb0z

[Permabear1993 @permabear1993]

@profplum99 @PtrPomorski @optionsniper420 be careful who you are talking too, this guy was Time's 2006 person of the year 😂

[Permabear1993 @permabear1993]

Just what healthcare needs, another middleman company skimming money off the top, wohoo!

[Patrick OShaughnessy @patrick_oshag]

My conversation with Zach Reitano (@ZReitano), co-founder and CEO of Ro

In four years he took Ro from a $250M run rate to over $3B, with about a third fewer people

Zach rarely does long conversations like this, so it was fun to talk about the details of what that took and everything happening in healthcare

Plus, his answer to the closing question is amazing

We discuss:
- the state of the US healthcare system
- refounding Ro
- strong attract, strong repel culture
- the notion of jobs over lives
- the power of celebrities in marketing
- the importance of tailwinds (AI, GLP-1s)
- trying to recreate his dad in software

Enjoy!

TIMESTAMPS
0:34 How Much Should We Spend on Health
4:59 The Patient Is Not the Customer
10:24 Inside Ro's Business
18:28 The Future of GLP-1s
25:46 Jobs Over Lives
30:02 The Turnaround
47:43 Where Healthcare Goes Next
1:00:27 Recreating His Dad in Software
1:06:33 Marketing Into Stigma
1:19:06 Zach's Hottest Take
1:20:58 The Kindest Thing

[Permabear1993 @permabear1993]

@Biotech2k1 i am not an active biotech investor (as a biologist, tbh i want to short everything as the claims are always beyond the data, thus i stay away as i have a bias) but if you ever want a pair of eyes, hit me up

[Permabear1993 @permabear1993]

@bshlgrs superintelligence is largely a useless semantic and I don't think there is yet any evidence AI has enough (or any) contextual awareness of reality, LLMs are stochastic machines

Not saying that AI can't "go rogue" and cause harm, as that does not require superintelligence

[Permabear1993 @permabear1993]

@LukewSavage I know nothing of the inside game, but I imagine its easier to convince a few dozen senators than it is hundreds of house members in a variety of caucuses

[Permabear1993 @permabear1993]

@OliviaHelenS looks like a rip of the 200 (apologize for hyperbole) similar programs that DTRA and DARPA have funded- what is unique here

[Permabear1993 @permabear1993]

@NickNemo17 haha fair