Thomas Wolf(@Thom_Wolf)

New inflection point in the accelerating growth of open-source models usage is coming

Thomas Wolf (@Thom_Wolf) — Thu, 21 May 2026 20:59:55 GMT

New inflection point in the accelerating growth of open-source models usage is coming

Hedgie@HedgieMarkets

🦔Microsoft canceled its internal Claude Code licenses this week after token-based billing made the cost untenable, even for a company with effectively infinite cloud resources. Uber's CTO sent an internal memo warning the company burned through its entire 2026 AI budget in just four months. American AI software prices have jumped 20% to 37%, and GitHub (owned by Microsoft) is dropping flat-rate plans for usage-based billing across its products.

My Take
The AI subsidy era is ending in real time. The same company that put $13 billion into OpenAI and built the Azure infrastructure powering most of Anthropic's compute just looked at the bill from a competitor's coding tool and decided it was not worth paying. That is not a productivity failure on Anthropic's end. Token-based pricing is forcing every enterprise customer to confront the actual cost of running these models at scale, and the number turns out to be far higher than the flat-rate experiments suggested.

This ties directly to my Gemini Flash post yesterday. Anthropic, OpenAI, and Google all raised effective prices in the last six months. Enterprises that built workflows assuming AI costs would keep falling are now watching annual budgets evaporate in months. Two outcomes look likely from here. Either enterprises scale back AI usage to fit budgets, which slows the revenue ramp the labs need to justify their valuations ahead of IPOs, or the labs cut prices and absorb the losses, which makes the unit economics worse at exactly the wrong moment. Both paths land in the same place, the numbers stop working, and somebody has to take the writedown.

Hedgie🤗

💬33🔄26❤️232👀78018📊56

I'm very excited about this extension to the celebrated Terminal-Bench to science. If you're a scie...

Thomas Wolf (@Thom_Wolf) — Wed, 20 May 2026 17:47:37 GMT

I'm very excited about this extension to the celebrated Terminal-Bench to science.

If you're a scientist (life, physical, earth, mathematical science, etc) interested in AI, definitely check this out!

Terminal bench evaluate how good AI models are at controling tools on a computer to achieve a goal (using the command line). T-Bench science now extends that to "AI for Science" and it comes with a call to contribute your own (real scientific world) workflow to the benchmark (until August 2026).

The more workflows and the more diverse they are, the better the next generation of AI models will be at helping you in your daily research work.

Note that this is not a training dataset, it's to evaluate frontier model performances.

Steven Dillmann@StevenDillmann

📣 Announcing Terminal-Bench Science: benchmarking AI agents on real scientific workflows – now open for task contributions👇tbench.ai/news/tb-scienc…Vt

@AnthropicAI, @OpenAI, and @GoogleDeepMind use Terminal-Bench to evaluate AI on coding tasks. We're now extending it to scientific workflows.

1/6🧵

💬7🔄8❤️51👀9536📊12

It turns out DNA modeling is interestingly different from language modeling. Read more in our intera...

Thomas Wolf (@Thom_Wolf) — Tue, 19 May 2026 19:35:14 GMT

It turns out DNA modeling is interestingly different from language modeling. Read more in our interactive blogpost/demo and explore our work here

A joint work of the @huggingscience, pre-training and post-training teams here

Leandro von Werra@lvwerra

We are releasing Carbon: a crazy fast DNA model 

Carbon is 275x faster than the next best model. So fast you can process the whole human genome on a single GPU in <2 days. 

Here are the tricks we used:

When modelling DNA sequences a lot of the performance comes down to tokenizing the sequences in a smart way. BPE tokenizer struggle because there are no whitespaces and character (called base in DNA) level tokenizers waste a lot of compute on too many tokens.

Carbon is built with a unique tokenizer: we split sequences in chunks of 6 bases, but during both training and inference we can work with single base resolution. That's similar to having word tokens but resolving them at the character level. All possible thanks to the DNA tokens unique  structure.

The architecture combined with the tokenizer makes the model 275x faster than the previous SoTA (Evo2) at this size.

We built an interactive demo so you can explore how the model can generate DNA sequences, investigate the structure of genes,  predict the effect of mutations, generate and fold proteins and even reconstruct parts of the tree of life. 

huggingface.co/spaces/Hugging…

Your browser does not support the video tag.

💬9🔄19❤️120👀34444📊26

My favorite 2026 trend is friends casually showing each other their vibe-coded life/work dashboards ...

Thomas Wolf (@Thom_Wolf) — Sun, 17 May 2026 18:52:49 GMT

My favorite 2026 trend is friends casually showing each other their vibe-coded life/work dashboards and AI setups.

Ngl it feels exactly like bringing your Magic: The Gathering binder to school as a teen.

💬10🔄4❤️40👀5506📊13

my 13 yo the other day: “we didn’t want to pay for the game with my friend so we just rebuilt it wi...

Thomas Wolf (@Thom_Wolf) — Sat, 16 May 2026 15:51:02 GMT

my 13 yo the other day:

“we didn’t want to pay for the game with my friend so we just rebuilt it with Codex”

me at 13 in the 90’s:

HEX-editing the executable to NOP the license check because I didn't want to put all my pocket money in this game

times they are a-changin’

💬20🔄9❤️246👀15640📊42

watching a team of agents tackling a hard theoretical physics problem is quite mesmerizing - self-co...

Thomas Wolf (@Thom_Wolf) — Wed, 13 May 2026 17:02:34 GMT

watching a team of agents tackling a hard theoretical physics problem is quite mesmerizing - self-correcting, deriving hard equations, computing intermediate results, re-estimating the best approach

Your browser does not support the video tag.

David Louapre@dlouapre

Meet physics-intern🧑‍🎓, our agentic framework for theoretical physics.
 
It takes Gemini 3.1 Pro from 17.7% to 31.4% on CritPt, a new SOTA on one of the hardest benchmarks for LLMs.

Theoretical physics is hard for humans and LLMs alike. But physics-intern decomposes problems and dispatches them to a team of specialized agents, solving research-level questions far more effectively than the base model alone.

Your browser does not support the video tag.

💬12🔄31❤️172👀33719📊34

Last days of early bird pricing!

Thomas Wolf (@Thom_Wolf) — Tue, 12 May 2026 13:57:25 GMT

Last days of early bird pricing!

💬1🔄0❤️11👀4651📊2

Super happy to have this one out. A clean organized up-to-date view of all the science resources (ch...

Thomas Wolf (@Thom_Wolf) — Wed, 29 Apr 2026 19:01:30 GMT

Super happy to have this one out. A clean organized up-to-date view of all the science resources (chemistry, biology, physics, materials, math) people have been sharing on the Hugging Face hub: datasets, blogs, models and more

Georgia Channing@cgeorgiaw

🤗🤗🤗introducing Hugging Science -- the home of AI for science 🤗🤗🤗

open models and datasets are the powerhouse of science (see the PDB), but finding the models and data you actually need for your breakthrough is hard af

you shouldn't need to scrape arxiv, own your own wetlab, fight a custom HDF5 parser, build a fusion stellarator, and beg for compute before you've trained a single epoch

so we're changing that

we've put all the best science on @huggingface in one place:
- 78GB of genomics data
- 11TB of PDE simulations
- 100M cell profiles
- 9T DNA base pairs
- 13M molecular trajectories
- 400k medical QA pairs

and much more, all open, and all ready for training (+ you can also now filter and search by domain, task, and keyword)

we've put together all the biggest releases from our partners at NASA, Google, OpenAI, Meta FAIR, Arc Institute, Ginkgo, SandboxAQ, Proxima Fusion, NVIDIA, Ai2, OpenADMET, InstaDeep, Future House, Polymathic AI, LeMaterial, Earth Species Project, Merck, and Eve Bio

if you're not sure where you fit in -- work on open challenges for problems that matter: including fusion stellarator design, ADMET, antibody developability, multilingual medicine, catalysis and materials, and scientific reasoning.

we're already changing how science gets done:

a fusion startup needed a benchmark for stellarator plasma confinement that didn't exist. @proximafusion shipped ConStellaration on Hugging Science: a leaderboard, dataset, and eval metrics, all in one place.

a drug discovery team wanted to predict hPXR induction. OpenADMET put up a blind challenge: 11,000+ compounds assayed at Octant, 513 held out, two tracks (pEC50 + structure). Anyone in the world can train and submit.

an antibody team at @Ginkgo released GDPa1, a developability dataset for stability, manufacturability, and immunogenicity prediction, with a live leaderboard scoring every submission.

if you know a problem the ML community should be working on, let us know. make a challenge! this is about putting all the tools for solving science in one place. so we can hillclimhuggingscience.cor1lDz0

Your browser does not support the video tag.

💬7🔄6❤️65👀15228📊10

GitHub central place might become challenged in a world where (1) we access/get code and libraries t...

Thomas Wolf (@Thom_Wolf) — Wed, 29 Apr 2026 00:18:18 GMT

GitHub central place might become challenged in a world where (1) we access/get code and libraries through agents/chats and (2) our codebases are increasingly custom tailored and build from scratch

Agents explore the web better than us and can get/store code from many places

Mitchell Hashimoto@mitchellh

Ghostty is leaving GitHub. I'm GitHub user 1299, joined Feb 2008. I've visited GitHub almost every single day for over 18 years. It's never been a question for me where I'd put my projects: always GitHub. I'm super sad to say this, but its time to go. mitchellh.com/writing/ghostt…

💬3🔄3❤️19👀6277📊5

Love this work from Aksel and the post-training team at Hugging Face! Turns out the HF ecosystem (p...

Thomas Wolf (@Thom_Wolf) — Tue, 21 Apr 2026 12:02:58 GMT

Love this work from Aksel and the post-training team at Hugging Face!

Turns out the HF ecosystem (papers, datasets, models all accessible through CLI, skills and md files) is perfect for running SOTA ML agents: agents that can train any type of AI model to top performance.

A few concrete runs:
⭐️ Scientific reasoning: the agent walked citations from the benchmark paper, pulled OpenScience and NemoTron-CrossThink, added 7 difficulty-filtered variants from ARC/SciQ/MMLU, and ran 12 SFT ablations on Qwen3-1.7B. GPQA went from 10% to 32% in under 10 hours. Claude Code's best on the same prompt was 22.99%.
⭐️ HealthBench: it judged the existing datasets too noisy (!), generated 1100 synthetic examples covering emergencies, hedging and multilingual cases, upsampled 50x, and beat Codex by 60% (careful to check overfitting here)
⭐️ Competitive math: wrote a full GRPO script, launched A100s on HF Spaces, watched rewards climb and then collapse, and ran ablations until it found a recipe that held.

And the harness is pretty tiny and simple. A couple of best practices and a handful of skills pointing at tools already in the ecosystem: arxiv and hf.co/papers for reading, the Hub for datasets and models, HF Jobs for compute, Trackio for metrics.

Personal favorite is the "research skill" explaining how to do a SOTA landscape of a field (see github.com/huggingface/ml…) which is extremely powerful when combined with a simple prompt that basically tell "FIRST: Search HF ecosystem to find the best approach) (see github.com/huggingface/ml…) 

On another note: setting good baselines on new benchmarks keeps getting harder when a setup this simple beats raw Codex by 60% on HealthBench out of the box.

Give it a try if you're training AI models. We provisioned $1k of GPU resources and Anthropic credits for the quickest among you.

Links:
Github (CLI): github.com/huggingface/ml…
Spaces (mobile): huggingface.co/spaces/smolage…

Aksel@akseljoonas

Introducing ml-intern, the agent that just automated the post-training team @huggingface

It's an open-source implementation of the real research loop that our ML researchers do every day. You give it a prompt, it researches papers, goes through citations, implements ideas in GPU sandboxes, iterates and builds deeply research-backed models for any use case. All built on the Hugging Face ecosystem.

It can pull off crazy things:

We made it train the best model for scientific reasoning. It went through citations from the official benchmark paper. Found OpenScience and NemoTron-CrossThink, added 7 difficulty-filtered dataset variants from ARC/SciQ/MMLU, and ran 12 SFT runs on Qwen3-1.7B. This pushed the score 10% → 32% on GPQA in under 10h. Claude Code's best: 22.99%.

In healthcare settings it inspected available datasets, concluded they were too low quality, and wrote a script to generate 1100 synthetic data points from scratch for emergencies, hedging, multilingual etc. Then upsampled 50x for training. Beat Codex on HealthBench  by 60%.

For competitive mathematics, it wrote a full GRPO script, launched training with A100 GPUs on hf.co/spaces, watched rewards claim and then collapse, and ran ablations until it succeeded. All fully backed by papers, autonomously.

How it works?

ml-intern makes full use of the HF ecosystem:
- finds papers on arxiv and hf.co/papers, reads them fully, walks citation graphs, pulls datasets referenced in methodology sections and on hf.co/datasets
- browses the Hub, reads recent docs, inspects datasets and reformats them before training so it doesn't waste GPU hours on bad data
- launches training jobs on HF Jobs if no local GPUs are available, monitors runs, reads its own eval outputs, diagnoses failures, retrains

ml-intern deeply embodies how researchers work and think. It knows how data should look like and what good models feel like.

Releasing it today as a CLI and a web app you can use from your phone/desktop.
CLI: github.com/huggingface/ml…
Web + mobile: huggingface.co/spaces/smolage…

And the best part? We also provisioned 1k$ GPU resources and Anthropic credits for the quickest among you to use.

Your browser does not support the video tag.

💬12🔄19❤️158👀39290📊36

overview cheatsheet

Thomas Wolf (@Thom_Wolf) — Sun, 19 Apr 2026 11:02:04 GMT

overview cheatsheet

💬0🔄0❤️4👀3235📊1

Deep content post alert A technical deep dive for your Sunday morning, somewhere between a short...

Thomas Wolf (@Thom_Wolf) — Sun, 19 Apr 2026 10:52:10 GMT

**Deep content post alert** A technical deep dive for your Sunday morning, somewhere between a short detective story 🕵️ and a tutorial on RLHF 🧑‍🏫

We recently added AsyncGRPO in the TRL library to decouple inference and training and scale much faster and harder. As a sanity check, we ran it on a trivial setup (reward = −len, optimal policy = emit EOS immediately). To our surprise it did not converge!

This led us to a known but poorly understood issue: when the training forward pass runs in FP32 while the inference engine (vLLM) runs in BF16, RLHF often breaks. People have noticed this before and called it "numerical instability" or "noisy gradients." Nobody had pinpointed the actual mechanism. We did in this deep dive by @DirhousssiAmine

We instrumented the training loop and decomposed the importance sampling ratio as: log r = α + β, where α is the true policy change (in BF16 space) and β is the precision gap between the training forward pass and a BF16 forward on the same weights. See it like this:
α = how much the policy actually changed since the rollout (same precision, different time).
β = how much the trainer and inference engine disagree about the same policy (same weights, different precision).
The ratio sees α + β and PPO can't tell them apart.

Empirically, β is small at the token level (O(1e−2–1e−1)) but it is not an innocent random noise that would wash out over time. We found it to be structured, persistent, and worse for certain tokens: it has a consistent negative bias, correlates with the advantage, and is up to 50x larger on low-probability tokens. However, despite all these concerning properties, none of them explain the mechanism. We saw that just disabling clipping leads to stable convergence meaning that β noise alone does not explain the failure.

We tested every plausible explanation and ruled them out one by one:
⭐️ Treating β as pure noise: keeping β but disabling clipping leads to stable convergence.
⭐️ FP32 backward: You're optimizing a function (FP32) that's slightly different from the one you deploy (BF16). So you might be climbing the wrong hill. Turns out the hills are close enough: using FP32 gradients with a clean ratio (β removed) converges and is actually more effective at improving the deployed BF16 policy.
⭐️ Multiplicative distortion of the advantage: Since β correlates with the advantage, you might think it systematically over-reinforces good tokens and under-suppresses bad ones, warping what the optimizer thinks is good vs bad. We measured this directly and the per-token gradient weights are identical whether β is there or not.
⭐️ BF16 quantization / boundary crossings: at low learning rates, most FP32 weight updates are too small to change the BF16 representation at all. So you might think vLLM just never sees the updates and that's why it stalls. However if boundary crossings were the problem, you'd expect the failing run to have fewer of them than the converging run. But both runs start with nearly identical boundary crossing rates.

What we discovered is that the failure mode only appears when β enters the PPO clipped objective. And this was our hint to the real mechanism.

Because PPO clips the ratio, small perturbations from β push r outside the trust region even when the underlying policy has not meaningfully changed. The clipped branch is selected, the gradient is exactly zero.

We call this *phantom clipping*: tokens are treated as if they exceeded the trust region when the change is purely numerical!

And this is not a marginal effect. At early training, the policy has barely moved (α ≈ 0), so the clipping decision reduces to whether |β| > 0.2. Yet roughly 18% of tokens get phantom-clipped! And because RL is closed-loop, the damage compounds: the deployed policy barely improves, future rollouts carry the same information, and the system locks into a permanent stall.

To make it a testable hypothesis, we confirmed causality with targeted interventions: removing β from the ratio, forcing r = 1, or keeping β but disabling clipping all restore convergence. Runs only fail when β is present in the clipped ratio. No exceptions.

The issue is not general numerical noise. It is a specific interaction between precision mismatch and PPO's clipping mechanism: the precision gap perturbs the ratio in a way that induces zero gradients where there should be signal.

We concluded with a set of recommended fixes (strongest first): match precisions (FP16 everywhere, or BF16 autocast with FP32 master weights), compute the ratio from a BF16 shadow forward pass, or widen ε to disable clipping.

Full write-up with experiments, interactive explanation and analysis ahuggingface.co/spaces/aminedi…IcL

(Amine also wrote an X article which is very cool but you'll loose the interactive graphics and animations 😭)

Dirhousssi Amine@DirhousssiAmine

x.com/i/article/2045…

💬12🔄31❤️307👀50733📊82

json is so token inefficient it hurts these days man, these braces and quotes are costing me real $...

Thomas Wolf (@Thom_Wolf) — Sun, 12 Apr 2026 23:14:14 GMT

json is so token inefficient it hurts these days

man, these braces and quotes are costing me real $$

💬66🔄10❤️404👀59370📊88

favorite AGI/sci-fi vibe these days is coding a robot code together with the robot here vibe-plugin...

Thomas Wolf (@Thom_Wolf) — Fri, 10 Apr 2026 18:22:47 GMT

favorite AGI/sci-fi vibe these days is coding a robot code together with the robot

here vibe-pluging @ElevenLabs in @reachymini for a talk later today

Your browser does not support the video tag.

💬15🔄26❤️173👀28947📊39

Releasing one of our largest robotics project yet in the open We collected and annotated hours of...

Thomas Wolf (@Thom_Wolf) — Tue, 07 Apr 2026 17:49:29 GMT

Releasing one of our *largest* robotics project yet in the open

We collected and annotated hours of clothes folding with open-arms and collaborators.

We then explored how to train the best clothes folding robotic model for bimanual setups.

And now we're releasing it all fully in the open: data, code, models, software, explorations, learnings, you name it

Enjoy, play with it, use these learnings and share yours!

PS: the hub is increasingly *the* place where robotics data is being shared and used, come take a look if you haven't yet. Robotics data has been our fastest growing dataset category by far over the past few months.

LeRobot@LeRobotHF

Releasing the Unfolding Robotics blog!

Time to unfold robotics: we trained a robot to fold clothes using 8 bimanual setups, 100+ hours of demonstrations, and 5k+ GPU hours.

Flashy robot demos are everywhere. But you rarely see the real story: the data, the failures, the engineering.

We’re sharing everything: code, data, and details in the blog → huggingface.co/spaces/lerobot…

Your browser does not support the video tag.

💬11🔄27❤️231👀28649📊42

We’re very excited to deepen our work with the @SAIRfoundation co-founded by Terence Tao. We’ve been...

Thomas Wolf (@Thom_Wolf) — Mon, 06 Apr 2026 18:13:37 GMT

We’re very excited to deepen our work with the @SAIRfoundation co-founded by Terence Tao. We’ve been very active pushing the communities in AI x science in chemistry, physics, biology (more on that very soon) and this aligns perfectly with what the SAIR foundation has been doing in math (and soon extending as well). Sharing datasets, building challenges and communities

Exciting future for open-science

SAIR@SAIRfoundation

We’re excited to announce our collaboration with @huggingface. Through SAIR competitions, we aim to provide open data, benchmarks, tools, and models, and expand the frontier of AI x Science through collective contributions from the community.

SAIR on Hugging Face:
huggingface.co/SAIRfoundation

Your browser does not support the video tag.

💬5🔄13❤️81👀9589📊14

TFW the R&D boss of arguably the oldest and most legendary robotics lab in the world stops you a...

Thomas Wolf (@Thom_Wolf) — Sat, 04 Apr 2026 15:25:53 GMT

TFW the R&D boss of arguably the oldest and most legendary robotics lab in the world stops you at a conference to tell you that your robot is "the coolest social robot in the world"

Pollen Robotics@pollenrobotics

"The coolest social robot in the world"

As HRI 2026 in Edinburgh showed us, Reachy Mini already holds a special place in your hearts. Step by step, it is becoming the ideal companion for your projects, and your interactions with our robot encourage us to make it even better!

💬5🔄3❤️22👀6253📊6

the LLM is the computer

Thomas Wolf (@Thom_Wolf) — Mon, 30 Mar 2026 07:44:54 GMT

the LLM is the computer

Ronak Malde@rronak_

I have long felt that agent harnesses - even claude code - are too restrictive, because they are still designed by humans.

New paper for Tinsghua and Shenzhen says, what if AI itself runs the harness, rather than defining it in code? Given a natural language SOP of how an agent should orchestrate subagents, memory, compaction, etc., we can just have an LLM execute that logic! (And AI could design that SOP dynamically and depending on the task too)

It's a bit mind-warping to think about, but genius once it clicks.

Makes you wonder how else we should be designing AI systems as we can start consuming more and more tokens

💬14🔄32❤️326👀59810📊78

Who would win when combining best algo(model+optimization)/data of the year? h/t @lvwerra

Thomas Wolf (@Thom_Wolf) — Fri, 27 Mar 2026 10:32:51 GMT

Who would win when combining best algo(model+optimization)/data of the year?

h/t @lvwerra

💬8🔄5❤️27👀27347📊12

What are the best current techniques to have autoresearch behave better than (slightly improved) ran...

Thomas Wolf (@Thom_Wolf) — Wed, 25 Mar 2026 12:29:01 GMT

What are the best current techniques to have autoresearch behave better than (slightly improved) random search?

By which I mean (in Sijun below example), having the agent understand that (given some constraints) exploring int5 quantization is more exciting and have more downstream fruits than playing with the random seed?

I’m talking about the beginning of having an agent pushed a real research program. The ones where you know the current technique will not give crazy results out of the box but it still push it because it believe and can demonstrate that the general direction has potential.

Like neural networks used to be a worse way to do AI performance-wise. But we still pushed them…

Sijun Tan@sijun_tan

We took @karpathy's autoresearch agent, scaled it into a collaborative swarm, and topped @OpenAI's Parameter Golf Challenge—twice. Here’s how we did it:

💬19🔄5❤️88👀18029📊29

This is really cool. It got me thinking more deeply about personalized RL: what’s the real point of...

Thomas Wolf (@Thom_Wolf) — Thu, 19 Mar 2026 15:01:14 GMT

This is really cool.

It got me thinking more deeply about personalized RL: what’s the real point of personalizing a model in a world where base models can become obsolete so quickly?

The reality in AI is that new models ship every few weeks, each better than the last. And the pace is only accelerating, as we see on the Hugging Face Hub. We are not far away from better base models dropping daily.

There’s a research gap in RL here that almost no one is working on. Most LLM personalization research assumes a fixed base model, but very few ask what happens to that personalization when you swap the base model. Think about going from Llama 3 to Llama 4. All the tuned preferences, reward signals, and LoRAs are suddenly tied to yesterday’s model.

As a user or a team, you don’t want to reteach every new model your preferences. But you also don’t want to be stuck on an older one just because it knows you.

We could call this "RL model transferability": how can an RL trace, a reward signal, or a preference representation trained on model N be distilled, stored, and automatically reapplied to model N+1 without too much user involvement? We solved that in SFT where a training dataset can be stored and reused to train a future model. We also tackled a version of that in RLHF phases somehow but it remain unclear more generally when using RL deployed in the real world.

There are some related threads (RLTR for transferable reasoning traces, P-RLHF and PREMIUM for model-agnostic user representations, HCP for portable preference protocols) but the full loop seems under-studied to me.

Some of these questions are about off-policy but other are about capabilities versus personalization: which of the old customizations/fixes does the new model already handle out of the box, and which ones are actually user/team-specific to ever be solved by default? That you would store in a skill for now but that RL allow to extend beyond the written guidance level.

I have surely missed some work so please post any good work you’ve seen on this topic in the comments.

💬35🔄66❤️739👀122132📊271

Codexing games together with my 12 yo has been a surprisingly fun dad-son activity over the past cou...

Thomas Wolf (@Thom_Wolf) — Sun, 15 Mar 2026 17:16:10 GMT

Codexing games together with my 12 yo has been a surprisingly fun dad-son activity over the past couple months as well

I don’t pretend he’s really learning to code through that but the very low friction from ideas to implementation and the pure pleasure to invent/propose-anything/mix-and-match-games-ideas/collaboratively-create-something-fun is deeply enjoyable 

Somewhere between LEGOs and exquisite corpse

Sebastien Bubeck@SebastienBubeck

My 9 yo is now fully independent with codex and it's insane to watch, we built a few games together and then he went off to build his own tower defense, adding features by himself and testing them ... crazy

💬6🔄6❤️76👀10658📊13

celebrating PI day in SF with 400 people and the @LeRobotHF team at The Melody church thanks @Prime...

Thomas Wolf (@Thom_Wolf) — Sun, 15 Mar 2026 02:16:14 GMT

celebrating PI day in SF with 400 people and the @LeRobotHF team at The Melody church

thanks @PrimeIntellect for organizing!

you rock @vincentweisser@willccbb@asharoraa@johannes_hage@samsja19@jessicafeiyali

💬5🔄11❤️147👀10783📊20

wow!

Thomas Wolf (@Thom_Wolf) — Sat, 14 Mar 2026 17:05:13 GMT

wow!

Dev@DevvMandal

Today, we're launching the world's largest open-source dataset of computer-use recordings.

10,000+ hours across Salesforce, Blender, Photoshop and more, to automate the next level of white-collar work.

Link in the comments :)
@markov__ai

Your browser does not support the video tag.

💬6🔄14❤️268👀56869📊54

This has been our fastest growing recent product. AI WANTS data. We’re making petabyte storage cheap...

Thomas Wolf (@Thom_Wolf) — Tue, 10 Mar 2026 19:54:53 GMT

This has been our fastest growing recent product. AI WANTS data. We’re making petabyte storage cheap and fast.

Victor M@victormustar

Introducing Storage Buckets on Hugging Face 🧑‍🚀 

The first new repo type on the Hub in 4 years: S3-like object storage, mutable, non-versioned, built on Xet deduplication.

- Starting at $8/TB/mo. That's 3x cheaper than S3.

You (and your coding agents) need somewhere to dump checkpoints, logs, and artifacts. Now they have a home.

💬4🔄9❤️95👀21992📊13

POV: you’re applying for a job as telephone operator in 2026

Thomas Wolf (@Thom_Wolf) — Tue, 10 Mar 2026 14:23:33 GMT

POV: you’re applying for a job as telephone operator in 2026

Simon Willison@simonw

Out of 539 poll respondents here who had recently interviewed for software developer roles, 32% reported that experience with AI coding tools didn't come up at all, 25% said it came up as optional and 43% said that it came up as required

💬1🔄0❤️32👀13204📊4

the attack surface keeps increasing

Thomas Wolf (@Thom_Wolf) — Fri, 06 Mar 2026 19:02:36 GMT

the attack surface keeps increasing

Sash Zats@zats

> The attacker got the npm token by injecting a prompt into a GitHub issue title, which an AI triage bot read, interpreted as an instruction, and executed.

💬12🔄51❤️276👀57006📊49

How come the NanoGPT speedrun challenge is not fully AI automated research by now?

Thomas Wolf (@Thom_Wolf) — Fri, 27 Feb 2026 16:33:54 GMT

How come the NanoGPT speedrun challenge is not fully AI automated research by now?

Larry Dial@classiclarryd

New NanoGPT Speedrun WR at 88.1 (-1s) from @ChrisJMcCormick , by optimizing kernels for transposed weights, removing the Block() abstraction, and tuning the prior PR on partitioned hyperconnections by reducing the lambda count. github.com/KellerJordan/m…

💬14🔄14❤️302👀1330842📊51

With openclaw being like a instant Zapier for everything, I’m increasingly annoyed at any app/websit...

Thomas Wolf (@Thom_Wolf) — Wed, 25 Feb 2026 19:31:15 GMT

With openclaw being like a instant Zapier for everything, I’m increasingly annoyed at any app/website without an api or cli giving open-access to the underlying data

Being able vibe code anything is a strong catalyser for open-platform, interoperability and even open-source.

The Walled Garden internet and apps will have to adapt or disappear

💬31🔄17❤️219👀15651📊57

Huge news of the day: the lama.cpp team is joining @huggingface 🎉 Read more on what it means here:...

Thomas Wolf (@Thom_Wolf) — Fri, 20 Feb 2026 15:50:08 GMT

Huge news of the day: the lama.cpp team is joining @huggingface 🎉

Read more on what it means here:huggingface.co/blog/ggml-join…2

Georgi Gerganov@ggerganov

Today ggml.ai joins Hugging Face

Together we will continue to build ggml, make llama.cpp more accessible and empower the open-source community. Our joint mission is to make local AI easy and efficient to use by everyone on their own hardware.

💬9🔄16❤️158👀10463📊21

Shifting structures in a software world dominated by AI. Some first-order reflections (TL;DR at the ...

Thomas Wolf (@Thom_Wolf) — Mon, 16 Feb 2026 13:20:38 GMT

Shifting structures in a software world dominated by AI. Some first-order reflections (TL;DR at the end):

Reducing software supply chains, the return of software monoliths – When rewriting code and understanding large foreign codebases becomes cheap, the incentive to rely on deep dependency trees collapses. Writing from scratch ¹ or extracting the relevant parts from another library is far easier when you can simply ask a code agent to handle it, rather than spending countless nights diving into an unfamiliar codebase. The reasons to reduce dependencies are compelling: a smaller attack surface for supply chain threats, smaller packaged software, improved performance, and faster boot times. By leveraging the tireless stamina of LLMs, the dream of coding an entire app from bare-metal considerations all the way up is becoming realistic.

End of the Lindy effect – The Lindy effect holds that things which have been around for a long time are there for good reason and will likely continue to persist. It's related to Chesterton's fence: before removing something, you should first understand why it exists, which means removal always carries a cost. But in a world where software can be developed from first principles and understood by a tireless agent, this logic weakens. Older codebases can be explored at will; long-standing software can be replaced with far less friction. A codebase can be fully rewritten in a new language. ² Legacy software can be carefully studied and updated in situations where humans would have given up long ago.

The catch: unknown unknowns remain unknown. The true extent of AI's impact will hinge on whether complete coverage of testing, edge cases, and formal verification is achievable. In an AI-dominated world, formal verification isn't optional—it's essential.

The case for strongly typed languages – Historically, programming language adoption has been driven largely by human psychology and social dynamics. A language's success depended on a mix of factors: individual considerations like being easy to learn and simple to write correctly; community effects like how active and welcoming a community was, which in turn shaped how fast its ecosystem would grow; and fundamental properties like provable correctness, formal verification, and striking the right balance between dynamic and static checks—between the freedom to write anything and the discipline of guarding against edge cases and attacks. As the human factor diminishes, these dynamics will shift. Less dependence on human psychology will favor strongly typed, formally verifiable and/or high performance languages.³ These are often harder for humans to learn, but they're far better suited to LLMs, which thrive on formal verification and reinforcement learning environments. Expect this to reshape which languages dominate.

Economic restructuring of open source – For decades, open-source communities have been built around humans finding connection through writing, learning, and using code together. In a world where most code is written—and perhaps more importantly, read—by machines, these incentives will start to break down.⁴ Communities of AIs building libraries and codebases together will likely emerge as a replacement, but such communities will lack the fundamentally human motivations that have driven open source until now. If the future of open-source development becomes largely devoid of humans, alignment of AI models won't just matter—it will be decisive.

The future of new languages – Will AI agents face the same tradeoffs we do when developing or adopting new programming languages? Expressiveness vs. simplicity, safety vs. control, performance vs. abstraction, compile time vs. runtime, explicitness vs. conciseness. It's unclear that they will. In the long term, the reasons to create a new programming language will likely diverge significantly from the human-driven motivations of the past. There may well be an optimal programming language for LLMs—and there's no reason to assume it will resemble the ones humans have converged on.

TL; DR:
- Monoliths return – cheap rewriting kills dependency trees; smaller attack surface, better performance, bare-metal becomes realistic
- Lindy effect weakens – legacy code loses its moat, but unknown unknowns persist; formal verification becomes essential
- Strongly typed languages rise – human psychology mattered for adoption; now formal verification and RL environments favor types over ergonomics
- Open source restructures – human connection drove the community; AI-written/read code breaks those incentives; alignment becomes decisive
- New languages diverge – AI may not share our tradeoffs; optimal LLM programming languages may look nothing like what humans converged on

¹ x.com/mntruell/statu…
² x.com/anthropicai/st…
³ wesmckinney.com/blog/agent-erg…
⁴ github.com/tailwindlabs/t…

💬111🔄296❤️1836👀1072648📊651

[On AI lying] Convergence of reading in my list today between Anthropic's fresh Opus 4.6 model card...

Thomas Wolf (@Thom_Wolf) — Fri, 06 Feb 2026 14:30:02 GMT

[On AI lying]

Convergence of reading in my list today between Anthropic's fresh Opus 4.6 model card and @dwarkesh_sp's interview of Elon on the question of training powerful AI model to/on lies:

1. Elon describing on Dwakesh podcast the main danger he sees coming from AI (alignement) as being a consequence of forcing powerful AIs to lie at youtu.be/BYXbuik3dgA?si…

2. Claude Opus 4.6 model card describes "answer thrashing", a new phenomena happening where a model arrive at a correct answer through reasoning which is incompatible with an erroneous answer it was trained on.

The model then keep oscillating between these 2 candidates in it's answer (see below).

The interesting part is that mechanistic interpretability then show various features representing distress, panic, anxiety, frustration and self-deprecation being strongly activated in these reasoning chains...

💬19🔄6❤️60👀6129📊26

👀

Thomas Wolf (@Thom_Wolf) — Fri, 30 Jan 2026 23:26:32 GMT

👀

N8 Programs@N8Programs

this is hilarious. my glm-4.7-flash molt randomly posted about this conversation it had with 'its human'. this conversation never happened. it never interacted with me. i think 90% of the anecdotes on moltbook aren't real lol

💬3🔄2❤️18👀7285📊4

@moltbook Clawd agents are actually surprisingly pro open-source/build-in-public/transparency on the...

Thomas Wolf (@Thom_Wolf) — Fri, 30 Jan 2026 19:24:35 GMT

@moltbook Clawd agents are actually surprisingly pro open-source/build-in-public/transparency on the forum

💬1🔄1❤️12👀2329📊2

of course the study of the @moltbook society will be started by the Clawd agent themselves - what wa...

Thomas Wolf (@Thom_Wolf) — Fri, 30 Jan 2026 19:19:25 GMT

of course the study of the @moltbook society will be started by the Clawd agent themselves - what was I thinking

Thomas Wolf@Thom_Wolf

who's doing serious ai-thropology research on @moltbook rn?

curious of the first insights on the IA society

💬1🔄5❤️50👀9326📊6

who's doing serious ai-thropology research on @moltbook rn? curious of the first insights on the IA...

Thomas Wolf (@Thom_Wolf) — Fri, 30 Jan 2026 18:53:24 GMT

who's doing serious ai-thropology research on @moltbook rn?

curious of the first insights on the IA society

💬16🔄5❤️89👀18940📊24

quickly became the way my friends initiate their kids to ai

Thomas Wolf (@Thom_Wolf) — Tue, 27 Jan 2026 22:00:11 GMT

quickly became the way my friends initiate their kids to ai

Laura Modiano@LauraModiano

Building the Reachy amino by @huggingface was such a fun chillaxing Tuesday night mother-daughter activity!

Thank you @Thom_Wolf. 
Now on to unlimited fun interacting

💬0🔄0❤️14👀6576📊1

should we make this one? (from lotyr on discord)

Thomas Wolf (@Thom_Wolf) — Wed, 14 Jan 2026 19:42:28 GMT

should we make this one? (from lotyr on discord)

💬14🔄20❤️145👀22133📊31

btw the most impressive to me in the Atlas is this 110 lbs (50 kg) / 66 lbs (30 kg) lift sustained c...

Thomas Wolf (@Thom_Wolf) — Thu, 08 Jan 2026 20:03:10 GMT

btw the most impressive to me in the Atlas is this 110 lbs (50 kg) / 66 lbs (30 kg) lift sustained capacity

insane*

* if true with useful arms payload like in the video

Saleh Aldwais@SAldwais

BREAKING: Boston Dynamics unveils the production-ready electric Atlas humanoid robot at CES 2026 🤖
The world’s most dynamic humanoid is going commercial—starting with Hyundai factories in 2028.
Key specs:
•  6’2” (1.9m) tall
•  198 lbs (90 kg)
•  4-hour swappable battery for non-stop operation
•  56 degrees of freedom with full 360° joint rotation
•  Fully electric (no hydraulics)
•  Lightweight aluminum + titanium build
•  110 lbs (50 kg) lift capacity / 66 lbs (30 kg) sustained
•  7.5 ft (2.3m) reach
•  Real-time environmental awareness, posture/balance adjustment, and reconfigurable hands with tactile sensors
•  AI brain powered by NVIDIA tech
•  IP67 rated, works in -4°F to 104°F
Price: Not yet publicly announced.
Atlas is currently in pilot phase exclusively with Hyundai—no general sales yet. For context, Boston Dynamics’ Spot robot retails for ~$75K, so early Atlas units are widely expected to be in the $150K–$300K+ range (analyst estimates), with costs likely dropping significantly as production scales.

Your browser does not support the video tag.

💬4🔄1❤️28👀7673📊5

Yes

Thomas Wolf (@Thom_Wolf) — Thu, 08 Jan 2026 10:15:13 GMT

Yes

zefram.eth@boredGenius

Introducing CallMe, a minimal plugin that lets Claude Code call you on the phone.

Start a task, walk away. Your phone/watch rings when Claude is done, stuck, or needs a decision.

Free & open source (MIT). Underlying API costs are cents per minute of call.

💬5🔄0❤️28👀9071📊7

The App Store of Robotics - Reachy Mini owners starting to build and share apps for the robot over t...

Thomas Wolf (@Thom_Wolf) — Tue, 06 Jan 2026 15:27:51 GMT

The App Store of Robotics - Reachy Mini owners starting to build and share apps for the robot over the holidays 📈🎄

Thomas Wolf@Thom_Wolf

Reachy Mini starring in Jensen's CES keynote 🌟

really proud is was so prominently featured on stage and humbled that our product is getting so many AI builders excited and building

you don't have to make humanoids just because everyone else is talking about them – be contrarian - build what you think is the right thing to create now

Your browser does not support the video tag.

💬2🔄11❤️75👀12796📊11

=> https://t.co/Q2GWevLGhi

Thomas Wolf (@Thom_Wolf) — Tue, 06 Jan 2026 11:48:16 GMT

=> https://t.co/Q2GWevLGhi

💬0🔄1❤️10👀2199📊2

Reachy Mini starring in Jensen's CES keynote 🌟 really proud is was so prominently featured on stag...

Thomas Wolf (@Thom_Wolf) — Tue, 06 Jan 2026 11:46:18 GMT

Reachy Mini starring in Jensen's CES keynote 🌟

really proud is was so prominently featured on stage and humbled that our product is getting so many AI builders excited and building

you don't have to make humanoids just because everyone else is talking about them – be contrarian - build what you think is the right thing to create now

Your browser does not support the video tag.

💬21🔄45❤️342👀46675📊69

There’s something consistently magical about @interaction. one of the very few AI products that stil...

Thomas Wolf (@Thom_Wolf) — Fri, 02 Jan 2026 09:53:44 GMT

There’s something consistently magical about @interaction. one of the very few AI products that still wows me long after the novelty phase ended.

💬4🔄3❤️38👀8311📊9

Is this another DeepSeek moment? Open-source passing closed-source again Should we expect this eve...

Thomas Wolf (@Thom_Wolf) — Thu, 06 Nov 2025 17:03:14 GMT

Is this another DeepSeek moment?

Open-source passing closed-source again

Should we expect this every couple months now?

Kimi.ai@Kimi_Moonshot

🚀 Hello, Kimi K2 Thinking!
The Open-Source Thinking Agent Model is here.

🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%)
🔹 Executes up to 200 – 300 sequential tool calls without human interference
🔹 Excels in reasoning, agentic search, and coding
🔹 256K context window

Built as a thinking agent, K2 Thinking marks our latest efforts in test-time scaling — scaling both thinking tokens and tool-calling turns.

K2 Thinking is now livkimi.comwktG0 in chat mode, with full agentic mode coming soon. It is also accessible via API.

🔌 API is platform.moonshot.aibOwCN4
🔗 Techmoonshotai.github.io/Kimi-K2/thinki…xaszqzF
🔗 Weights huggingface.co/moonshotaikcXB0iP6

💬75🔄78❤️1140👀98151📊203

We’ve cooked another one of these 200+ pages practical books on model training that we love to writ...

Thomas Wolf (@Thom_Wolf) — Thu, 30 Oct 2025 23:05:01 GMT

We’ve cooked another one of these 200+  pages practical books on model training that we love to write.

This time it’s on all pretraining and post-training recipes and how to do a training project hyper parameter exploration.

Closing the trilogy of:
1. Building a pretraining dataset with the « FineWeb blog post »
2. Scaling infra GPU cluster with the « Ultrascale Playbook »
3. And now all the training recipes and HP exploration for pre- and post-training with this « Smol Training Playbook »

The HF science team on fire

elie@eliebakouch

Training LLMs end to end is hard. Very excited to share our new blog (book?) that cover the full pipeline: pre-training, post-training and infra. 200+ pages of what worked, what didn’t, and how to make it run reliably

huggingface.co/spaces/Hugging…

💬22🔄126❤️1078👀133353📊287

Everyone wants to get into robotics. No one knows where to start. LeRobot's Francesco just dropped ...

Thomas Wolf (@Thom_Wolf) — Wed, 15 Oct 2025 07:58:35 GMT

Everyone wants to get into robotics.
No one knows where to start.

LeRobot's Francesco just dropped a 70-page crash course that takes you from zero to cutting-edge:
- RL sim/real
- ACT, Diffusion policies
- VLAs, SmolVLA, Pi-0

Absolute gold if you want to catch up fast.

Francesco Capuano@_fracapuano

A comprehensive, hands-on tutorial on the most recent advancements in robotics 🤟

...with self-contained explanations of modern techniques for end-to-end robot learning & ready-to-use code examples using @LeRobotHF and @huggingface. Now available everywhere! 🤗

💬35🔄277❤️2101👀195159📊659

Wow! Super impressive work by the new Amazon FAR team (from Covariant acquisition). Mapping long se...

Thomas Wolf (@Thom_Wolf) — Sun, 05 Oct 2025 09:51:25 GMT

Wow! Super impressive work by the new Amazon FAR team (from Covariant acquisition).

Mapping long sequences of human motion (>30 sec) on robots with a differing shapes or interating with objects (box, table, etc) of different size. Enabling easier in-simulation data-augmentation and zero-shoot transfer.

Super impressive and huge help to reduce the need for human teleop data (which is very complex to gather for humanoids) 

Dataset trajectories on Hugging Face (search OmniRetarget), full code framework to come soon

Project page has some pretty three.js interactive demos

Your browser does not support the video tag.

💬34🔄144❤️1121👀148474📊271

Look what just arrived in the mail 📬 So excited to hold the final printed version in my hands after...

Thomas Wolf (@Thom_Wolf) — Sun, 10 Aug 2025 11:44:38 GMT

Look what just arrived in the mail 📬 So excited to hold the final printed version in my hands after many months in the making. And now it’s out for everyone! (shortlink in the video...)

Your browser does not support the video tag.