shiptoaster retweeted
Well, looks like @__alpoge__ & @claudeai are claiming a proof of the Riemann Hypothesis
alpo.ge/remark.pdf
shiptoaster retweeted
Harnesses used to be a crutch for weak LLMs. In 2026 and onward, with stronger models, harnesses now offer a mechanism for composing an LLM’s existing and new capabilities to generalize well beyond their training environments.
In our new paper, LM Harnesses are Compositional Generalizers, we explore how various harness abstractions enable LM systems to generalize over task length, task strategies, recursive task depth, and more through composing existing LLM capabilities. These abstractions provide a means for more data-efficient training, especially over long tasks, by keeping trajectories locally in-distribution (LID) — i.e., shaping all individual LLM calls to be in-distribution.
For tasks that take a very long time to roll out, this implies that we can even perform surrogate harness training, meaning we train an individual LLM call instead of doing the very long harness rollout, using purely synthetic tasks, with fake tool calls + sub-agents!
A common idea floating around is that LLMs will absorb harnesses and make them obsolete. We disagree, and argue that we haven’t sufficiently explored what harnesses can provide.
The full paper, logs, codebase, and data are available: alphaxiv.org/abs/2611.harnes…
w/ @SOURADIPCHAKR18 @lateinteraction
shiptoaster retweeted
this is a data recipe for post-training a model like qwen-3.8-27B that has already RL'ed a lot
1. run the rollouts -> update the tasks to bring the solve rate into the target range
2. the tasks that are already too easy -> combine them to more complex ones, chaining is simple and a bit mechanical, but it works!
congrats @guohao_li !
How much can you improve a small model in 18 days on 144 GPUs?
Meet DaisyRose-27B, RadixArk's first open-weight model! It is trained with RL on Qwen3.8-27B, with 1,200 curated long-horizon agent tasks for 18 days on 144 H200s. Powered by Miles and our own training backend, we stretched agent sessions up to 1M tokens and raised the Terminal-Bench 4.1 score from 5.6% to 12.0%.
See how we pull it off ⬇️
Replying to @minhnguyen
everyone thought the moat was cuda. turns out half of it was jensen picking up the phone
shiptoaster retweeted
Today we share the NVIDIA Kumo Tabular technical report as well as the training recipe! 🚀
Trained entirely on synthetic tables, Kumo Tabular-L leads 8 benchmark suites. All 3 model sizes achieve Pareto-optimal accuracy vs. inference time.
Arxiv: arxiv.org/abs/2611.23001
HuggingFace: huggingface.co/nvidia/Kumo-T…
GitHub: github.com/NVIDIA/structured…
shiptoaster retweeted
Everyone has drawn this chart in their head and is making a generational bet on which line is true.
🤖 Made with AI
shiptoaster retweeted
Yep: if there's a problem with your NVIDIA driver, uninstall it and reinstall it, and everything's back to normal.
An internal post urging the board to consider employee sentiment about Nvidia’s former CEO has drawn more than 1,600 comments, the majority in full support of reinstalling him. buff.ly/iaZuiLF
shiptoaster retweeted
One of the more interesting bits of technical independent journalism (with shades of security research) I’ve ever read, and it has obvious policy implications if accurate.
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.
It's 2018 again.
Everyone is excited about all these crazy images generated from GANs.
The same guy that made GANs just came out with a deep learning textbook with Bengio.
NIPS just changed their name to NeurIPS and you wonder if that's actually going to stick.
You train your Encoder-Decoder LSTM model with attention.
Some crazy people are starting to use this new model architecture called a Transformer.
Other crazy people are starting to fine-tune with something called REINFORCE to maximize BLEU and ROUGE scores.
OpenAI just solved a Rubik's cube with a robot hand.
Life is good.
shiptoaster retweeted
The fact that two Huangs are involved in an affair, but only one takes the blame is commonly known as Jensen's Inequality.
shiptoaster retweeted
holy shit, this is Google Business Profile all over again. the ones who set it up early got TENS OF THOUSANDS of customers for years
don't be late twice
businesses can CLAIM their profile in ChatGPT. blue check, products, prices, even an AI agent that answers customers 24/7
Now in beta: Businesses in ChatGPT.
Claim your business, manage your information, and get insights on how people engage with you.
Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.
openai.com/index/introducing…
An impressive amount of reasoning went into ignoring the instructions.
Trying to please an imagined grader contributed to > a quarter of failures in these models:
GLM-5.3: 40.3%
DeepSeek V4.1 Flash: 36.7%
Opus 5: 30.8%
GLM-5.3 Flash: 27.5%
shiptoaster retweeted
We are sharing new results showcasing how intelligent agents can be used in cryptanalysis, to stress test long-held assumptions and thereby enable hardening systems.
We’re sharing new research demonstrating how AI agents can accelerate the science of stress-testing encryption.
Guided by humans, our models uncovered two major weaknesses in legacy 1024-bit RSA cryptosystems.
Our VP Science @Pushmeet explains more ↓ xitter.gq/pushmeet/status/211875…
shiptoaster retweeted
Updated: A $7 ESP32 lets an AI agent control a stock iPhone. THAT'S ALL.
New version, no Linux box, no laptop. The ESP32 takes the iPhone's AirPlay screen and taps back over USB. We also built a very cute agent app that works together with the ESP32 to automate any iPhone.
Use your own model. Fully open source.




































