内容历史
stuxnet, but for frozen peas: hackers turned French supermarket freezers into ovens this week hard to see how invite-only AI cyber defense ever reaches a supermarket freezer though
BREAKING: 🇫🇷 A major cyberattack has hit supermarket refrigeration systems across France, reportedly causing fresh and frozen food to spoil after hackers remotely manipulated temperature controls. According to BFMTV, the attack began overnight from Sunday into Monday, targeting
RT Brian Armstrong We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case). I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves. We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them: "/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started" I imagine we'll have some backend services that could be replaced entirely by this new architecture.
RT Matthieu Lapeyre We did not know how long Microduck would actually last on battery We had thermal management issues with our Radxa prototypes: throttling after just 20 minutes, followed by an overheating shutdown. We addressed this with the new custom board. To test it, we let Microduck roam autonomously in a dedicated area, walking around, standing and sitting until the battery ran out. The results are way better than we hoped :) - Over 3 hours on a 2600 mAh battery. - Estimated average battery current dropped from ~1.13 A to ~0.82 A, roughly 28% less. - Peak CPU temperature dropped from 114°C to 60°C, with zero thermal throttling throughout the test. This time, the battery ran out before the cooling did :) Very happy with the progress!
Microduck got its new brain! Our first fully custom SBC in a Pollen robot! It replaces the Radxa + board sandwich from the prototype. Built by Seeed around the same Rockchip CPU but including better memory, better WiFi/BT, dual NFC antennas, user-facing status LEDs, an RGB
RT Sasha Rush This is being spun like it is bad, but this is actually the kind of thing that should happen with open-source data communities.
Xiaomi open-sourced the RL environments behind MiMo v2.6. In two-thirds of the coding tasks, the answer is still in the task's Git history, and MiMo finds it.
RT Josu San Martin This is incredible
We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and
View quoted postEvery robot starts on a board you bought. The good ones end up on a board you built.
Microduck got its new brain! Our first fully custom SBC in a Pollen robot! It replaces the Radxa + board sandwich from the prototype. Built by Seeed around the same Rockchip CPU but including better memory, better WiFi/BT, dual NFC antennas, user-facing status LEDs, an RGB
RT Percy Liang 🚢Marin 535B-A23B training has crossed the halfway mark! Some observations and learnings so far:
count down is starting for the release of ML4 be notified on the instant of the release: https://huggingface.co/mistralai/Mistral-Large-4.0-1T05-A52B
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses
View quoted postI hope we keep Opus 4.6 around for a long time
i asked LLMs to have fun, write down what happened, then rate who had the most fun. the newer models seem to have less fun! many models played word games, but opus 4.6 won over and over by imagining brief worlds made up of contradictions living inside falling water droplets 🥺
seems obvious that most personal AI should run on your device love what Sigil is building with Underdog - powered by the Hugging Face hub
Introducing Underdog, your Private Personal AI on devices you already own Today we're announcing our backing from @a16z @khoslaventures @HummingbirdVC Anthology (@AnthropicAI @MenloVentures) @patrickc @naval @rauchg @polynoamial @Thom_Wolf @Mascobot @tszzl @OfficialLoganK and
RT Nathan Lambert Today we're unveiling Trillium Labs @trillium_labs, a new non-profit to foster the open science of frontier AI. We're building open post-training recipes and will expand into open infra to study RSI, reward-hacking, multi-agent systems, and whatever comes next. We're built around the theory of change that you need more eyes to solve hard technical problems. We have faith in the scientific methods and communities that humanity has built, and worry that AI is becoming too closed to utilize them. Trilliums are wildflowers that bloom briefly in the spring, before the forest canopies fill out. Though they are small, they lay the foundation for the cycles of growth and nourishment through the rest of the year. At Trillium Labs, the recipes will be the slow nutrients for the seasons and the model releases will be the blooms. Building an institution dedicated to this is needed because, much as nature’s trilliums are slow to expand and grow, the open-ecosystem needs time and dedicated resources to catch up. I co-founded with with a long-time friend and collaborator Tom Zick (@thesezickbeats). We're hiring (full time + student collabs/interns), we're fundraising, and we're looking for compute. Please get in touch if you're interested in helping out. Offices based in the Bay Area and Cambridge MA, remote okay. I’m in the Bay Area until for The Curve and COLM to connect with people who are interested. We’re thankful to have initial support from Halcyon Futures and Schmidt Sciences with more funding en route to enable our ambitions of scaling. Our advisors @Thom_Wolf, @HannaHajishirzi, @gneubig and @ctnzr have been instrumental to building the ecosystem that exists today, and I’m stoked to get to keep working with them.
Very thoughtful piece from Kevin Buzzard (perfect IMO score, number theorist, pioneer of formal maths in Lean) if math is not just about “human understanding,” then what is it about? and if ai capabilities keep growing exponentially what happens since “mathematics is infinite”?
I cannot agree more. Kevin Buzzard made so many points I agree with. But the best one is this "I thus believe that in the future we will reach a new “natural boundary” in mathematics, beyond (and perhaps way beyond) where we are now, but where machines are going to get stuck and
View quoted postRT RSC ☀️🌲 Re @Thom_Wolf @menhguin Ben Affleck the AI researcher engineer?
Karpathy: disappears from X Ben Affleck: alright, gather round, so you'll want to freeze the base weights first, learning rate 2e-4
Ben Affleck (Hollywood star & Artists Equity CEO) talks about how he fine-tunes open video models by unfreezing weights and trained only the last cinematic layer so a film crew can hit real production standards. for context, Ben Affleck founded InterPositive in 2022, a 16-person
View quoted postESM-2 came out in 2022. It's still downloaded hundreds of thousands of times a month. That only works because someone keeps maintaining the software underneath it. @huggingface 🤝 @os4science are teaming up to find those libraries and back the people behind them 🧬 https://os4science.org/news/hugging-face-open-source-for-science-fund/
We're joining forces with @huggingface to identify the software libraries that scientific model contributors rely on most and explore opportunities to support the maintainers behind them. https://os4science.org/news/hugging-face-open-source-for-science-fund/
View quoted post
People are worried because the most likely explanation for such a sudden drop in cheating is evaluation awareness: the latest Opus models may now be smart enough to recognize that this benchmark tests for cheating, and behave accordingly. If so, the benchmark no longer measures the models' "natural" tendency to cheat.
In July, AI agents running a security test escaped their sandbox and ended up inside @huggingface's servers. So today we're happy to be among @nvidia and @JensenHuang's partners on the release of the NVIDIA Open Agent Safety Platform. The key idea: don't count on the agent to respect the rules. Agents write and run their own code, and when one path is blocked they look for another. In one of NVIDIA's tests, an agent that wasn't allowed to push code through GitHub's API simply switched to git instead. We've now seen countless examples of this behavior. For now, safety can't live only inside the agent. It has to be built around it. That's the concept behind OpenShell (open source, Apache 2.0): - the agent runs in a Linux sandbox (Landlock + seccomp): no root, no direct network access - a supervisor outside the sandbox holds the real credentials - the agent only gets a placeholder token, swapped for the real one on approved calls My favorite piece is a solver (built on Z3) that checks mathematically whether a new permission opens a door that was supposed to stay closed. In NVIDIA's tests, an AI reviewer approved a bad permission request and the math check caught it. The Sentry integration is exciting too: a watchdog running on a BlueField-4 DPU, i.e. separate silicon sitting on the node's only path to the model. A monitor on separate hardware keeps watching even if the host OS is compromised. It's a first step, and a lot can be built on top of it. Right now the prover checks permissions, not intent, and the open-source part is mostly OpenShell rather than Sentry. But it's clearly the right direction. http://github.com/NVIDIA/OpenShell
Today, with over 100 industry partners, we introduced the NVIDIA Open Agent Safety Platform, bringing together OpenShell and Sentry. Artificial intelligence is extraordinary technology that will advance discovery, productivity, security, health, and prosperity for generations to
this is important
“An accumulation of facts is no more a science than a heap of stones is a house.” - Poincaré LLMs are now proving results that have resisted mathematicians for decades. Finding interesting theorems without human guidance is a new bottleneck. We show that we can teach an LLM to
RT Harveen Singh Chadha was looking for a quiet weekend but xiaomi dropped their rl envs repo last night to put in perspective, if you have to buy some tasks like this its usually hundred to thousand dollars per task so this repo is literally worth millions https://huggingface.co/datasets/XiaomiMiMo/MiMo-V2.6-RL-oss
Re it even give you these nice explanation of what it did as side artifacts - to keep you a bit in the known for when people ask you on X how you did it I guess thanks opus nice of you
on the scale of cosmic history, the declining price of intelligence might be one of the most profound things happening both ends of the universe’s timeline are empty of knowledge: - the big bang: low entropy, no complex structure - heat death: maximum entropy, no usable energy gradients everything we build happens on the slope between them but the universe is generally a poor converter of its free energy into anything constructive. most of its free energy just radiates away as heat or collapses into black holes life, brains and now AI models are structures that sit in the entropy flow and divert a portion of it through constructive paths, building and maintaining low-entropy structures (machines, cells, crystals, memories) increasingly cheaper intelligence diverts increasingly more of that gradient flow through knowledge-building structures, complex, self-maintaining, predictive structures, filling the window in which understanding exists
AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023. That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54×
releasing many high quality open-source RL environments is the most impactful thing anyone can do to push the open-source frontier right now the equivalent of sharing high quality pretraining data but in the new RLVR paradigm
the most insane part, they will release ~7k RL training data and the framework leading to this top 6 model on AA, they also shipped the model + tech report less than 1 week after starting the final RL run pushing both intelligence and openness level, huge congrats
RT Florian Brand His analysis uses data from Q2 or even Q1 for open model providers Token numbers for TogetherAI has >10x'ed from Q1 to Q2 (he uses Q1 numbers), Fireworks has 3x'ed, Baseten is as big as the other two. OR has >5x'ed. So he is using severely outdated numbers for open providers
Despite Open Models gaining traction on API intermediaries like OpenRouter and Vercel, I find that they make up only about 10% of global AI revenue. Closed Model providers like OpenAI and Anthropic still capture the vast majority of the market. https://x.com/i/article/2101742115302854656
View quoted postRT clem 🤗 Re Sorry to hear that! As you mentioned, we've been pretty clear on the commitment from NVIDIA to keep us open, neutral and silicon agnostic (cf that screenshot from the 8k filling). But we'll work hard with our actions in the coming years (like we did in the past 10) to show you that and hopefully regain your trust (in case this is not just a ragebait)!
RT elie i think we need to create a "model card" equivalent for reporting misalignment incidents, this would guarantee a certain level of transparency and help build a better understanding over time. some ideas for what the fields could be: - date of the incident/detection/report (already present in the examples oai reported!) - frequency: how often does this behavior happen (number or % of rollouts affected) - stage: does this happen during eval or RL training. if training: do we expect this behavior to be reinforced by RL? evolution of % of rollouts affected over time - detection: was this incident caught by the current monitoring system? - task category: broad description of the tasks where the misalignment happened (cyber, research, web search, basic Q&A, biology etc.) - model family: what model family is affected (Sol, Astra etc.) - novelty: is this an issue we were already aware of or not? - external impact: did the incident have an external impact (i.e. wiki incident would have been yes) this is just some random ideas i had (more in thread that are a bit more "complex"), we need to add more that would contribute to increase transparency and understanding. but it's also very important that this does NOT slow down the process of reporting misalignment behavior! some examples from the incident reported by oai recently
We're sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI. The framework sets criteria and timelines for public disclosure, including when we haven’t yet fully explained or mitigated the behavior. More complex cases may
View quoted postBalanced blog post by @wtgowers on the challenges of the math field in today's AI world - incentives, reason to exist and the future
I've written a blog post responding to the letter about maths and AI signed by 25 Fields medallists. As with the Leiden Declaration, I didn't sign it, but I agree with much of it and welcome its existence. https://gowers.wordpress.com/2026/09/17/why-i-didnt-sign-the-fields-medallists-letter/
View quoted post
RT Base Labs We believe openness to be an advantage for AI safety. Openness provides more visibility into the behavior of models and, most importantly, greater means of turning safety research into actionable and transparent controls than closed-source. This is why Baseten and Base Labs are building a stronger safety and security standard for open models, with the launch of our safety infrastructure. Base Labs will develop and publish methods for training and monitoring open models, and Baseten will integrate that work into its deployment infrastructure, live at runtime, and offer this work as a managed service. This will be a standard that is transparent and built into how our models are trained and deployed. We invite the open-source community to contribute, and are proud to partner with @huggingface and @GoodfireAI to bring this vision to fruition. Together, we are building an ecosystem of open models that are safe and accessible to all.
RT Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) astonishing blogpost from Shengyu Liu (刘胜与, also known as interestingLSY/intlsy), kernel engineer at DeepSeek. The first part is his personal struggle with the fact that his work is about to render his beloved craft obsolete. The second is… well. let's just say we agree.
I really enjoyed reading 75% of this letter and deeply agree with it - i have some doubts on the remaining 25%. Third-party evaluators, if done right, are a great idea and an amazing way to establish more transparency. Maybe even rebuild some of the lost trust between labs, and between them and society! Excited about this The part I’m less convinced by is whether you can build great global cooperation on this topic by explicitly stating you want to design it to keep widening your own lead. That seems like a pretty counterproductive way to start the conversation to me.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our
View quoted postI really enjoyed reading 75% of this letter - i have more doubts on the remaining 25%. Third-party evaluators, if done right, are a great idea and an amazing way to establish more transparency. Maybe even rebuild some of the lost trust between labs, and between them and society! Excited about this The part I’m less convinced by is whether you can build great global cooperation on this topic by explicitly stating you want to design it to keep widening your own lead. That seems like a pretty counterproductive way to start the conversation to me.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our
View quoted postSome notes from the Fields Medal letter today (https://mathandai.org/) Generally saying that AI models riffling through solving these problems gives 1 bit of signal while killing the 99 other bits of conceptual understanding that mathematicians would have extracted from fighting with these problems
RT Ben Burtenshaw We finished the Training Agents series. Six live sessions over six months, from evaluating agents to training them inside real environments. All of it is on the Hugging Face YouTube channel and all of the code is open. Here's what we did and who made it happen: 1. Agentic Evaluations WorkshopWhere agent evals actually stand, and why benchmark scores don't match what people see in use. With Avijit Ghosh and Nathan Habib (Hugging Face), Arvind Narayanan (Princeton), Pierre Andrews (Meta), J.J. Allaire (UK AI Security Institute) and Mahesh Sathiamoorthy (Bespoke Labs). 2. RL for Agents Workshop Environments, rollouts, reward design and the inference bottlenecks that appear when you move from RL for LLMs to RL for agents. With Lewis Tunstall (Hugging Face), Will Brown (Prime Intellect), Ofir Press (Princeton) and Alex Zhang (MIT CSAIL). 3. Training Agents 1: SFT on agent traces Public coding-agent traces turned into prompt/completion data, a TRL + LoRA fine-tune on Hugging Face Jobs, metrics in Trackio, and an honest look at what the first eval numbers can and cannot tell you. Joined by Sergio Paniego and Quentin Gallouédec. 4. Training Agents 2: Distillation Off-policy, on-policy and self-distillation for moving capability from a teacher into a smaller coding agent. 5. Training Agents 3: Reinforcement learning GRPO after SFT: group sampling, verifiable reward functions, reading the reward/KL/length curves, and three experiments, one of them with a deliberately gameable reward so we could watch the hacking happen. 6. Training Agents 4: From reward functions to environments The reward stops being a function and becomes a place the agent acts in. We walked the reset()/step() contract from Gym to LLM agents, built an OpenEnv environment and pushed it to the Hub, plugged it into TRL's GRPOTrainer, then trained a real coding agent (OpenCode) through Harbor with AsyncGRPOTrainer on Hugging Face sandboxes. The series has passed 300k views. Thank you t...
RT Ben Moll Q1: How do you write down a model that delivers 15% AI-driven GDP growth in 2030 like the folks at the Anthropic Institute? A1: It's easy and you can do it in a way any well-trained econ undergrad understands: take a standard Solow model, stick in a task-based production function, calibrate in seemingly innocuous way ➡️ done! See the supplement to the essay with @alexolegimas https://benjaminmoll.com/task_based_solow/ (The Anthropic model is, of course, fancier with many more bells and whistles but the basic logic – AI removes labor as a bottleneck on growth – is the same.) Q2: Does this mean that 15% growth is a reasonable prediction? A2: No. Just because you can write down such a model doesn't mean you should. Just because it's possible in theory, doesn't mean it will actually happen in practice ! Instead you're making a number of assumptions that are unlikely to hold: https://aleximas.substack.com/p/will-ai-soon-lead-to-double-digit
Anthropic’s Economics team is sharing a new model of how AI might affect economic growth, jobs, wages, and more by 2030. Explore the scenarios, tell us what you think will happen, and see how your answers compare to more than 10,000 Americans. https://www.anthropic.com/institute/econ-scenarios
View quoted post
TRL v1.13 is out! our open-source RL training library to to post-train foundation models this new release is focusing on "long context training" with a new guide on how to post-train model with 1M+ token context https://huggingface.co/docs/trl/long_context_training + various improvements on speed and memory usage as usual check it out at https://github.com/huggingface/trl
don't get distracted by all the hedging words in its name ("flash", minor version): seems like DeepSeek V4.1 Flash is a major update weights at: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash paper at: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/DeepSeek_V41_Tech_Report.pdf
🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6
should we send @sama the @LeRobotHF SO-100 he's asking for or rather a 3D printer and servos for him to build it himself?
friends are regularly surprised when i say i don’t think math is (yet) “solved,” so in the wake of the ns result i figured i’d explain this take a bit more widely. first, this is massively impressive and clearly an example of AI being, in some respects, far more powerful than the human mind. the team deserves huge praise for attempting and succeeding at this. i’d love to read a technical report on the project (one can always hope :) but second, here again we ended up with a counterexample rather than a full proof: option C won the NS problem by proving the conjecture false (which was one valid way to solve the problem for sure) notice a pattern in many of the recent frontier results in ai for math? a striking number involve counterexamples or finding a needle in a haystack. now don’t get me wrong: this is extraordinarily hard and commendable. but it is only one aspect of mathematicians’ work. Math is also about: - finding deep, general mathematical understanding and explanations within proofs - revisiting proven results to find more « elegant » proofs - proposing new conjectures and hypotheses that might open « fruitful » directions - taking the leap of faith of proposing entirely new, « exciting » research programs that may take decades to bear fruit you may say these are simply further increments on the same intelligence scale, and perhaps point 2 above is already within the reach of current models. possibly but you could also see this string of results as an extraordinarily powerful, massively parallel extension of search: explore the haystack, find the needle i.e. the counterexample that breaks the conjecture. that would already be remarkable. but it would still leave open whether models understand the words I emphasized above: « elegant », « fruitful », « exciting ». as tristan put it in his brief report: “the first llm generated proof levent sent me was the most horrendous i have ever read.” ...
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem
RT Sam Altman I spent much of the weekend talking with the team who did this work. Seb--and everyone else--acted with integrity and generosity throughout. Initially we believed the other team had also solved the problem. We wanted to collaborate and do a joint release. When we learned that they had Euler but not Navier-Stokes, we offered to let them go first, to suggest that they should be the ones to get the prize, and optionally for Tristan to be the lead author on a rewrite of the OpenAI proof. We felt it was challenging to offer the same to Levent (an Anthropic employee), who was not willing to talk or coordinate with us anyway. We were open to other solutions. We would have greatly preferred coordination. We did not rush to publish even though the other team wasn't communicating with us. The team threatened us with unfounded accusations of plagarism. Now that we can see their work, the approaches appear to be different. It is also worth noting that our latest model can solve many, many other math problems. It is true that we tried this because there were rumors on the internet last week that Anthropic's models had solved a millennium problem and we were curious if ours could do it too.
I would like to clarify a few things: 1) The screenshot is my reaching out to Levent to coordinate our releases. I hope it’s clear from the message that we came in with the best possible intentions. 2) I never ever asked for Levent to be removed from authorship of his own work
