How is your permission fatigue? Mine was all green at the third try :D
llmgame.scalex.dev/
1
61
Anna Nachesa 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.
Finally, got to watch The Discovery of Heaven movie, featuring Steven Fry who's totally awesome. It is somewhat simplified in comparison to the book of Harry Mulisch, but it wasn't possible otherwise. Beautiful movie.
1
105
Anna Nachesa retweeted
New University of Illinois paper finds that when AI decides if a research idea is new enough to work on, it often says no to ideas that human scientists later publish.
AI scientist pipelines run a novelty check to decide which generated ideas get compute. These checks judge a whole proposal at once, though most real papers build on some earlier contributions.
They built ClaimNov, which splits a proposal into a few claims, matches each against claims from prior papers, and gives a judge the claim-by-claim evidence.
A deployed check caught all 380 disguised copies but also rejected 96% of ideas researchers later published. ClaimNov caught 97% of copies while rejecting only 1.5% of those ideas.
If your pipeline filters ideas for novelty, score claims separately and keep proposals that build on 1 known result instead of discarding them.
There might be a dedicated place in purgatory for folks who use deodorant sprays in public trains.
1
4
112
Anna Nachesa retweeted
I took about 50,000 photos through a 14" telescope last night to produce this: A hyper-detailed photo of the full moon. The fidelity of the photo was great enough to allow me to greatly enhance the color, revealing a hidden world of geological diversity.
"Do not reply" might well be a key phrase of our time. #random_thoughts
1
85
Asked Claude (Opus Medium 5.5) to write a short story about how across generations people are frustrated that new tech deprives next generations of humans from basic survival skills. I didn't specify how it should end. That's what it wrote, no edits from me except the prompt. docs.google.com/document/d/1…
76
Anna Nachesa retweeted
Massive. New Harvard study finds that AI-written paper reviews now beat the best human peer reviews on every measure the authors tested.
Peer review is slow and often misses errors. Earlier studies concluded AI should only assist human reviewers, not replace them.
They used 134 reviews from rapidPeer, a Harvard service that pays expert reviewers well. Claude Opus 4.8 reviewed the same 1124 papers, and GPT-5.5 plus 3 former journal editors judged the pairs blind.
GPT-5.5 picked the AI review in all 132 decided pairs, and the editors picked it 87% of the time. Only 5.5% of the AI's claims about the paper were wrong, versus 15.9% for the human reviews.
Teams that depend on expert critique can now let AI do the first full review pass, and keep a human for the final decision.
Anna Nachesa retweeted
So basically, today PHP got two critical bugs that let anyone run code on your server without even logging in, which is bad, and then the details leaked before the patch was even out, which is worse, and it works on WordPress and MediaWiki and Magento and Joomla and Laravel, and Censys says like 1.5 million servers are running a version with no fix, so the NSA was like "take your system offline," and the FBI was like "take your system offline," and CISA and the UK and CERT-EU were all like "take your system offline," and then the Prime Minister of Japan, the actual Prime Minister, posted at 5:55 in the morning telling 48.8 million people to take their system offline, so everyone took their system offline, like the Dutch cybersecurity agency took its own website offline, and the Dutch parliament went offline, and the Port of Rotterdam went offline, and Finland and France and Denmark started pulling government sites, and Wikipedia went down, and Archive.org went down, and Asana and GOG and Something Awful went down, and the University of Vienna's webmail went down for some reason, and that's just PHP. Because GitLab also dropped a CVSS 10.0, which is the highest score they give out, for a bug that's been there since 2014, so that's every self-hosted GitLab for twelve years, and then LiteLLM 1.107.0 got hijacked and shipped a credential stealer and a remote access trojan in every single wheel and the Docker images, including latest, and you didn't even have to import it, just installing it was enough, and it's the second time this year, and then Railway paused every deployment on the entire platform because someone got into the environment variables, so now everyone has to rotate every secret they've ever stored there, and TiDB Cloud said someone was walking around inside with stolen credentials for basically a month and an extortion group says they took 10.6 terabytes, and Zeabur got breached for the second time in two months and is just shutting down, like the whole company, and Node.js patched remote code execution and a TLS bypass, and CISA says Plex is being actively exploited, and someone took over 300 Tor exit relays to steal SSH passwords, and SafePay ransomware has hit over 600 critical infrastructure orgs, and India's biggest bank filed a data loss incident, and Japan has been in an official state of emergency in cyberspace since October 9, which honestly explains the 5:55 AM post. Oh, and eMudhra, a certificate authority that's issued over 60 million certificates, whose entire job is proving websites are who they say they are, was supposed to be using private keys that are random numbers over a hundred digits long, and one of them was 4. Just 4. And the others were like 66 and 237. Anyway, it's Wednesday.
Anna Nachesa retweeted
Replying to @fiveoutofnine
Anna Nachesa retweeted
To give context on age:
Grace Hopper began computing at 38, completed the first compiler at 46, helped shape COBOL at 53, kept developing COBOL for the Navy in her 70s, retired from the Navy at 80, & then became a consultant for the Digital Equipment Corporation.
v/@cooperx86
Couldn't believe this happened: GHC has been cancelled this year. anitab.org/press-release
77
Anna Nachesa retweeted
The first self-driving cars in Switzerland started a few days ago. It's a tiny pilot project called "iamo" in collab with SBB (swiss rail) that brings people to the next train station. Very small, only 4 cars and with safety driver sitting behind the wheel. Uses tech from Chinese company WeRide.
Hopefully it goes well and "warms up" the public and regulators so that many other self-driving cos will follow soon!
Anna Nachesa retweeted
You can literally type a sentence and watch a Transformer process it.
This interactive explainer runs a GPT-style model in the browser and lets you follow the whole path:
tokenization → embeddings → Q/K/V → attention → MLP → logits → next-token probabilities
You can inspect attention maps, intermediate values, and change temperature, top-k and top-p to see how generation changes.
If Transformers still feel like boxes connected by arrows, this is worth opening.
poloclub.github.io/transform…
After having been to Greenland, I think this lady made a great choice given the circumstances. businessinsider.com/day-in-t…
70
Anna Nachesa retweeted
So... In the last few days I have had my agents write:
1. A word processing app in which I will write future books. It replaces Scrivener.
2. A script writing app in which I will write future video scripts. It replaces FinalDraft.
3. Crapper, a tool that measures and records CRAP statistics for projects. The tool handles many different languages, and adding new ones is easy.
4. Mutator, a tool that runs mutation tests and records the results. The tool handles many different languages and adding new ones is easy.
5. Dryer, a tool that runs duplication analysis on a project. The tools handles many different language and adding new ones is easy.
6. UML-Viewer, a tool that displays the structure of a project in a UMLish diagram that shows modules and dependencies and highlights dependency cycles. Clicking on a module displays it's contents; which may be deeper UML diagrams or, at the leaves, code. The tools color codes the modules to reflect their CRAP and Mutation scores, and as you drill down you can see those scores for each module and function. The tool understands many different languages, and adding new languages is easy.
I recon that, last year, this would have been well over a man-year of effort.
All these tools are in github.com/unclebob. I advise you not to use them directly. Instead, point your agents at them and use them as templates to build the tools that are customized to your needs. The licenses are all BSD Zero, meaning that there are no strings. Do what you like.






















