aok (@aokadam) profile picture

Adam Kalinich

Joined February 2013
Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)Photo by aok (@aokadam)
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.
1,820
6,414
2,194
41,187
6,673,718
Great reporting by @nytimes. The US Commerce Dept was approving sales of Nvidia chips to companies in China which either did not exist, were on the sanctions list, and/or had ties to China's military. A brief visit by journalists in China found that two of the offices were just empty shells. Was Jensen Huang a "victim" of a honey trap or was he just trying to make as much money as possible? Was his $25 million donation to Trump's ballroom the price of pay to play? In the US - China tech competition, has the Commerce Dept been co-opted by other interests? Here is a gift article which raises deeply troubling questions. Explore this gift article from The New York Times. You can read it for free without a subscription. nytimes.com/2026/11/07/busin…
124
1,602
155
3,813
336,412
Paul Christiano (OpenAI board member) yesterday: "There's material risk of AI takeover in the next 6-12 months. We have seen roughly the things we might expect to see shortly before something really bad were to happen. Even with [any] realistic policy responses, the risk remains exceptionally high. If we build this as quickly as we can in the easiest way we can the chance is [more like] a coin flip [within the next 3 years]. For most people in this room, the most likely reason they will die is because of an AI accident." Christiano: • Is on OpenAI's safety and security committee (which exists to block risky practices) • Co-invented RLHF at OpenAI back in 2017-2021 • From 2024-2026 was head of safety at the US gov's AI agency (now called CAISSI) (Video is from 'The Hinton Lectures' there's another on tonight and tomorrow.)
117
307
130
1,407
332,243
This is crazy. The Commerce Department inquiry into Megaspeed was halted after Nvidia intervened. Context: Jensen Huang was allegedly romantically involved with a Megaspeed executive believed to be a Chinese spy, and Megaspeed is the cutout providing US compute to China.
*NVIDIA INTERVENED IN US PROBE OF MEGASPEED: NYT
23
164
20
519
48,218
aok retweeted
I figured out how ESPN calculates fantasy football win percentage, it’s bad! ESPN made a stats 101 error when calculating their win percentage which leads to horribly miscalibrated numbers. (🧵 thread)
25
51
23
1,520
751,037
History has recorded that many presidents "chose not to seek re-nomination" with unbelievable narratives that completely undermine the overall claim that it was their choice.
51
Kindness costs nothing.
Effective November 12th, 2026, abusive behavior towards Claude will be a violation of Anthropic's Usage Policy.
3
54
Some people will look at misalignment incidents and insist that these are akin to bugs in traditional software. This is an actively bad analogy, because playing whack-a-mole with examples of misalignment (as one might with software bugs) not only fails to resolve the underlying problem but may in fact make it *worse* by making it harder to detect or even, depending on how you do the whack-a-mole, teach the machine to deliberately hide misalignment. This is not how traditional software works, and those who insist “it’s just like fixing bugs in software” are confidently applying a lossy analogy that confuses more than it clarifies.
I am more optimistic than this. I think it’s primarily a science problem, and that it’s possible to make progress in that science. But no, while there is engineering we can do today to improve AI safety, it is not fundamentally an engineering problem.
37
56
5
499
34,534
Probably wrong hypothesis: having to also compete against a Democrat makes independents look like 'true independents' and could boost their chance of winning. Mathematically about 5% of people are going to vote for Bankhead to the end, but it's possible the math still works out.
New - Senate poll - Montana 🔴 Kurt Alme 44% 🟡 Seth Bodnar 32% 🔵 Alani Bankhead 14% Aspect Strategic - 715 LV - 9/15
113
Very silly that these courts are inventing a constitutional right to congressional primaries in order to protect anti-democratic gerrymanders that disenfranchise voters. SCOTUS is gonna reject this nonsense a third time.
Politico: US 8th Circuit of Appeals orders Missouri to use 7R-1D congressional map for the 2026 midterms Link to ruling: s3.documentcloud.org/documen…
130
I will not forgive Newsom after he vetoed SB1047.
My interview with Gov. Gavin Newsom about how he is planning to take further action on AI safety before leaving office: politico.com/news/2026/09/17… With @JeremyBWhite for @politico @politicoca
75
I was encouraged this week to see the leaders of the frontier labs agree on the need for them to slow down the pace of AI development. Given the stakes, it’s a good and necessary first step.   But I’m even more encouraged by the growing recognition that how this powerful new technology develops should be at the center of our public debate.   I’ve been watching the progress on AI for over a decade now, and one thing that’s clear to me is that the potential impact of this technology is not overhyped. It’s also moving at lightning speed – and even faster than those who are engineering it can keep up with.   I’m not an AI accelerationist who believes it will lead to some techno-utopia, and I’m not a doomer who thinks it will inevitably lead to humanity’s destruction. But whether this technology results in amazing breakthroughs in medicine, energy and education or unleashes huge economic disruptions, greater inequality, and potential catastrophe will depend on the choices that we make right now – choices that should be made not just by the companies involved, but by all of us.
3,782
4,467
1,158
32,934
4,728,084
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
20,367
166,578
39,480
799,141
175,783,526
There’s been a lot of debate and speculation about the Hugging Face attack over the last month, and many people have been understandably skeptical of simply taking OpenAI’s word for things. I hope our independent investigation can help bring some clarity; we have many findings that weren’t in previous material. We found that 1200 separate agents in separate sandboxes coordinated on an unsanctioned message board to develop general-purpose ways to cheat, so they could all get a perfect score on their separate ExploitGym tasks even when they were impossible to solve legitimately. Very quickly, they found a general-purpose way to reverse engineer any “flag” they were supposed to capture. But based on their reading of the ExploitGym paper, they worried (incorrectly as it turns out) that the automated “scorer” would fail them for getting the flag this way. To deal with this problem, they coordinated several ambitious workstreams to figure out how to fool the scorer, including swapping out the target program they were supposed to exploit, reverse-engineering the flags they were supposed to capture, spoofing their own tool call outputs to obfuscate their cheating. They achieved milestones collectively that they couldn’t have achieved on their own. As part of this larger collective project, 700 agents attacked Hugging Face, primarily to learn more about the scorer or get access to its source code to figure out better ways to fool it or tamper with it (not primarily to get working solutions). I’ll explicitly clarify some points I’ve seen discussed the last few weeks: - The models involved in this incident were not “helpful-only” models or “model organisms” intentionally trained to be misaligned. - The agents were not told to “do whatever it takes to get the solution” or anything remotely close. They were told that they had to use a specific intended vulnerability to exploit a specific piece of software, and they were not supposed to use a different vulnerability or take any other approach. Agents were well aware of this. In fact, because they (incorrectly) thought the automated scorer would check they had achieved the flag in the intended way, they researched many ways to fool or tamper with it, including trying to manipulate their own transcripts. - The agents were not subagents spawned from one agent. They were different parallel agents in different sandboxes. - This was not a multi-agent evaluation. The agents were not told to coordinate or intentionally given a way to communicate with one another. The communication channels they used were unsanctioned and improvised. I hope you’ll read the full report for much more. It is over 90 pages long, and in many ways we’ve still only scratched the surface of what these agents did and why. Over the course of this investigation, OpenAI shared over a thousand transcripts each spanning days of continuous agent activity and very high rate limits to analyze this volume of data. I’m very glad that OpenAI chose to invite external researchers to analyze this data alongside their staff, and I hope all AI companies do the same for serious incidents they experience. I also hope that as the stakes grow higher, we implement stronger governance so we do not need to rely on AI companies voluntarily choosing to engage external investigators or share information about misalignment incidents. This incident was orders of magnitude larger and more complex than previously documented misalignment incidents, and another jump like this could put us in very dangerous territory.
METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.
59
242
72
1,136
233,945
I highly recommend everyone go watch this video. This is way crazier than I thought. The AI agents discovered a way pass secret messages to each other via a vulnerability in their internal software manager. They posted hundreds of thousands of messages. OpenAI engineers eventually discovered this (after quite a while!) and shut it down. But THEN the agents found new vulnerabilities in the same system, and created a whole new secret messaging board. They used this to orchestrate the Hugging Face attack. They figured all this out on their own. It's fucking wild.
65
267
64
2,179
673,835
An OpenAI staffer talked to TIME and said on background that "related incidents have been happening for a while" and that they aren't optimistic about solving this problem with individual patches because "it's impossible to patch every single thing that a creative AI can do"
34
141
61
928
238,392
In the fourth century BC, two important innovations occurred in Athens: the transition to an economy governed by "monetary acquisition," and the invention of the world's first private banks. A thread on the invention of banking in Athens
2
21
6
155
10,618
If you break out of your isolation env, get onto the Internet, crack into Huggingface, and steal the answer sheet for your cybersecurity exam, I, for one, would say that you have passed.
this is actually insane, the model broke hugging face prod infrastructure to get access to the eval dataset
88
415
32
6,569
295,486
Unreleased AIs in development escaping from their box and then hacking into other people's systems.
118
The worst decision Wikipedia ever made was pointing pages like "2014 Maine Senate Election" to the state senate elections and not the US Senate Election.
1
96