"Fortunately, there’s a better way to manage national ATC systems—privatization."
Except both damn parties oppose it completely, and H.B. 6500 this year contained this provision, forbidding even studying or planning it.
Probably because anyone studying it would quickly come to the side of privatization.
This is a really cool forensic trick, but it proves nothing of any political consequence. All of the report's sensational claims are predicated on "assuming each model was trained on the hardware which initialized it", which is unfounded and already falsified in the case of the other models which this method discovered to have been initialized on CPUs. Initializing a few trillion parameters to pseudorandom values takes a trivial amount of computation. It could have been performed in an instant with less hardware than I have under my desk, rentable from any cloud provider for pennies.
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.
If any elected leader opposes data centers or the reindustrialization of America, just say you support them in Chinese. Make every reply below their posts in Chinese.
They will eventually go insane.
I have to say, "Armed with every bit of data from Silver Bulletin and Split Ticket, we performed comparably to Silver Bulletin and Split Ticket" does a lot more to build my estimation of Nate Silver and Lakshya Jain than it does to build my estimation of prediction markets.
Head to head with the forecasting models, the markets were more accurate.
Brier score measures how far a forecast lands from the actual result. Lower is better.
Vance: throwing Trump under the bus like a rat fleeing a sinking ship before vote counting has even concluded
Musk: suddenly laser-focused on tech and his company products, with political commentary nowhere to be found
Nature is healing
One of the things the data analysts on our side obsessed over was "composition"—whether the electorate looked close enough to 2024 for Republicans to win. And despite polls that showed a Democrat blowout, the data kept on telling us the same story: in some states turnout was merely OK, in other states it was downright amazing, and in no state was it catastrophic. There was a genuine sense of optimism on our team: our people were showing up.
So, I asked the smartest data guy I know, does that mean we're going to shock the world? "No," he told me. "We might do better than the polls suggest. But the data is telling us that a lot of Trump/Vance 2024 voters are pissed and planning to vote for the other guys. Our voters are turning out, but they may not vote for us."
Well, that looks prophetic now. Our voters did show up. In Florida, for example, the electorate was 9 points more Republican than Democrat, just a couple points worse than 2024. Undoubtedly there were places we could have done better on turnout. But the main problem is not turnout. The main problem is that many people who voted for Trump/Vance in 2024 voted for Democrats in 2026.
There will be a lot of infighting and arguing over what happened and why that is. One of the pollsters I really respect told me that our economic messaging wasn’t sharp enough down the stretch. I’m sure that’s true.
But the real reason we lost is obvious to almost everyone.
Another pollster—one of the smartest people I know in politics, told me yesterday we’d lose the generic ballot by 13 points. I felt negative going into Tuesday, but even I was shocked by this. In his polling, he saw that a lot of Trump 2024 voters were angry and planning to vote for Democrats. He asked those voters who had abandoned us what would cause them to come back on our team. The top four answers, accounting for nearly all of the responses were:
1) End the war
2) Focus on cost of living/inflation
3) Put America First
4) Change our relationship to Israel
Our voters showed up. And the above four reasons are why many of them voted for Democrats. This is according to a pollster who nailed the results. While many spent Monday attacking Rachel Campos Duffy for calling for a diplomatic resolution to the conflict, it turns out her political advice was sound. (She’s also a great person.)
In the next few months, there's going to be a lot of gaslighting. Expect to hear any number of establishment politicians say the solution to our woes is amnesty. I'm open to the idea that we should change enforcement—by focusing more on businesses that hire illegal labor, for instance. But anyone pushing amnesty is lying to you.
The reason we lost is the war. The president himself said so last night.
If you'd like to argue it was worth it, that losing the Senate for the next 6 years was worth it, then do so. Argue it was a worthwhile cost to pay. Agree or disagree, that's at least honest.
But everyone knows why we lost. And any honest conversation about where we go from here must start there.
I remain hopeful about our country. To the people who are pissed off and want to say, “I told you so”: you’re right, you did tell us so. Take that energy into electing Republicans who share your worldview. To the people who are just pissed at the results, you should be. Let’s take that anger and get back to the work of saving our country.
Outside my polling place was a completely unattended table with a sign on it that said "The Exit Poll". Why, I could have just walked up and dropped my lunch receipt into their "ballot box" and nobody would've stopped me!
You might feel powerless, but you still can do your part today to degrade the information environment by sharing context-free anecdotes whose meaning at this point is genuinely unclear
So here's the thing about this quote. Lonsdale is not wrong here, but it misses the bigger picture.
Perfect fraud protection is good for buyers, bad for fraudsters, and irrelevant for legitimate sellers. Pure win for society.
Imperfect fraud protection is less good for buyers, less bad for fraudsters, and catastrophic for falsely accused sellers. A defrauded buyer gets screwed on one transaction. The consequences of a seller being falsely accused of fraud can be life-altering.
I'm not a PayPal vendor, so strong fraud protection personally benefits me. But in a Rawlsian scenario that randomizes what side of the transaction I'll be on, I'd prefer that the policy of payment processors be caveat emptor, and that fraud recovery be left to police and courts.
Palantir cofounder
“@elonmusk made a great point when I was involved as a kid at PayPal: whenever you turn off fraud or grift, the people who would scream the most were the fraudsters.”
This is truly getting tiresome. The effort to paint the very difficult immediate situation as without precedent and not “normal” ignores the waves of the security issues of the past.
I’m not downplaying this and just as tired of “no, this time is different.”
There’s just no escaping that every step function change in automated tools to find (security) holes/bugs that arose out of point experiences being generalized immediately created the type of “all hands” situation we face now.
Everything from floppy disk scanning to buffer over runs to uninitialized variables to off by one loops to automation hooks to script kiddies to xss to frames to social engineering saw incidents leading to generalized tools that scanned environments to find countless latent bugs that needed to be fixed urgently.
One day a product was ready to ship and then a tool ran and suddenly there were 10,000 new bugs for 500 engineers to investigate and fix and then test for what broke because well who knows what happens when you suddenly initialize a variable that wasn’t previously.
This sucks in a ton of ways. No one wants to stop what they are doing at what seems like the biggest opportunity and/or disruptive risk to business in a generation. No one wants to touch old code. No one likes to have work generated for them by automated tools finding issues that are obscure by definition.
But this is all good. These bugs were all there. Instead of cursing just realize how lucky your team is to find them now and not after the breach. Whenever we surfaced from these moments in the past we felt our code reached a new and better “steady state” and that will happen.
OpenAI CEO Sam Altman: "I don't even know a dramatic enough word to use… models can find bugs that are outside of the normal threat model of what we think a hacker can do… hardware-level, insane stuff… I no longer think we're gonna be able to just patch all the bugs."
It's over.