Thomas Wolf
Co-founder and chief science officer of Hugging Face; writes about open models and where he thinks the field is wrong.
Thomas Wolf did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake
Latest x.com
What they believe
-
A unified pan-European startup structure (EU-INC) is the best thing Europe could do to catch up in AI.
EU–INC is the single best thing Europe could do to catch-up in the AI race
bsky.appSaid 27 Sept 2025This on korrents
-
Low-tech AI, small cheap models plugged into existing systems, will bring surprisingly rapid societal change.
This shift toward low-tech AI will lead to surprisingly rapid societal changes.
thomwolf.ioSaid 14 May 2025This on korrents
The bold line is korrents’ wording; the quote is theirs. These 2: what they wrote before what they said aloud, one per source, newest first.
Feed
As its own page →Hiding
4 November
30 October
29 October
28 October
26 October
23 October
22 October
21 October
16 October
22 September
4 July
27 September 2025
-
korrents.com
A unified pan-European startup structure (EU-INC) is the best thing Europe could do to catch up in AI.Their words
EU–INC is the single best thing Europe could do to catch-up in the AI race
17 September 2025
10 September 2025
9 September 2025
7 September 2025
9 July 2025
8 July 2025
29 May 2025
19 May 2025
16 May 2025
14 May 2025
From one piece 🎠 Low Tech AI 2 beliefs · thomwolf.io
-
Their words
Extrapolating along these two lines means a fixed level of intelligence gets cheaper every year, fit increasingly more affordable hardware.
-
korrents.com
Low-tech AI, small cheap models plugged into existing systems, will bring surprisingly rapid societal change.Their words
This shift toward low-tech AI will lead to surprisingly rapid societal changes.
28 April 2025
- Bluesky
- GitHubsesame-explorations
22 April 2025
14 April 2025
4 April 2025
31 March 2025
26 February 2025
From one piece 🔭 The Einstein AI model 3 beliefs · thomwolf.io
-
korrents.com
Scientific AI needs a B student who questions what everyone missed, not an A+ student who answers every question.Their words
We don't need an A+ student who can answer every question with general knowledge. We need a B student who sees and questions what everyone else missed.
-
Their words
We're currently building very obedient students, not revolutionaries. This is perfect for today’s main goal in the field of creating great assistants and overly compliant helpers. But until we find a way to incentivize them to question their knowledge and propose ideas that potentially go against past training data, they won't give us scientific revolutions yet.
-
korrents.com
On current trends AI will give us a country of yes-men on servers, not a compressed 21st century of discovery.Their words
What we'll actually get, in my opinion, is “a country of yes-men on servers” (if we just continue on current trends)
31 January 2025
From one piece 🐳 Some notes on "DeepSeek and export control" 2 beliefs · thomwolf.io
-
korrents.com
Open-source AI's most fundamental long-term virtue is resilience, which makes it essential for safety.Their words
Open-source has many advantages like shared training costs, tunability, control, ownership, privacy but one of its most fundamental virtue in the long term –as AI becomes deeply embedded in our world– will likely be its strong resilience.
-
korrents.com
Open-weights DeepSeek makes the closed-source US-China AI arms-race argument artificial.Their words
having one of the model freely widely available for download and with detailed scientific report renders the whole « close-source arm-race competition » argument artificial and unconvincing in my opinion.
27 January 2025
-
korrents.com
AI intelligence will be commoditized; the remaining moat is the quality of the product.Their words
Intelligence will be commoditized! The remaining moat will be the quality of the product and how it integrate with co-factors.
15 January 2025
-
korrents.com
AI model evaluations should move to agentic evals.Their words
All evals should move to agentic evals in 2025 in my opinion.
25 June 2024
19 March 2024
Nothing matches.
Beliefs
Korrents What they believe 11 beliefs — each backed by an exact quote.
Each is a — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.
Recent
A unified pan-European startup structure (EU-INC) is the best thing Europe could do to catch up in AI.
EU–INC is the single best thing Europe could do to catch-up in the AI race
@thomwolf.bsky.social on Bluesky Said 27 Sept 2025
Low-tech AI, small cheap models plugged into existing systems, will bring surprisingly rapid societal change.
This shift toward low-tech AI will lead to surprisingly rapid societal changes.
🎠 Low Tech AI Said 14 May 2025
A fixed level of AI intelligence gets cheaper every year.
Extrapolating along these two lines means a fixed level of intelligence gets cheaper every year, fit increasingly more affordable hardware.
🎠 Low Tech AI Said 14 May 2025
Show 8 more
On current trends AI will give us a country of yes-men on servers, not a compressed 21st century of discovery.
What we'll actually get, in my opinion, is “a country of yes-men on servers” (if we just continue on current trends)
🔭 The Einstein AI model Said 26 Feb 2025
Today's LLMs are obedient students and will not produce scientific revolutions until they learn to question their training data.
We're currently building very obedient students, not revolutionaries. This is perfect for today’s main goal in the field of creating great assistants and overly compliant helpers. But until we find a way to incentivize them to question their knowledge and propose ideas that potentially go against past training data, they won't give us scientific revolutions yet.
🔭 The Einstein AI model Said 26 Feb 2025
Scientific AI needs a B student who questions what everyone missed, not an A+ student who answers every question.
We don't need an A+ student who can answer every question with general knowledge. We need a B student who sees and questions what everyone else missed.
🔭 The Einstein AI model Said 26 Feb 2025
Open-weights DeepSeek makes the closed-source US-China AI arms-race argument artificial.
having one of the model freely widely available for download and with detailed scientific report renders the whole « close-source arm-race competition » argument artificial and unconvincing in my opinion.
🐳 Some notes on "DeepSeek and export control" Said 31 Jan 2025
Open-source AI's most fundamental long-term virtue is resilience, which makes it essential for safety.
Open-source has many advantages like shared training costs, tunability, control, ownership, privacy but one of its most fundamental virtue in the long term –as AI becomes deeply embedded in our world– will likely be its strong resilience.
🐳 Some notes on "DeepSeek and export control" Said 31 Jan 2025
AI intelligence will be commoditized; the remaining moat is the quality of the product.
Intelligence will be commoditized! The remaining moat will be the quality of the product and how it integrate with co-factors.
@thomwolf.bsky.social on Bluesky Said 27 Jan 2025
AI model evaluations should move to agentic evals.
All evals should move to agentic evals in 2025 in my opinion.
@thomwolf.bsky.social on Bluesky Said 15 Jan 2025
Google missed a chance to retake the lead in AI by not making Gemini open source.
Google really did miss an opportunity to leapfrog both OpenAI and Meta and regain a place as 👑 of AI by making Gemini open source
@thomwolf.bsky.social on Bluesky Said 6 Dec 2023
What is a korrent?
A korrent is a belief a person has stated in their own words: one sentence stating the claim, backed by a quote and a source, kept at korrents.com.
Under a name here, the quoted block is what they actually said. The korrent beneath it is the claim those words support, in korrents' wording — tap it to see the record, its source, and who else holds it.
Nobody here wrote their own korrents. They are compiled from public statements, and a person can change their mind, which is recorded too.
Recommends rcmnd
rcmnd What they recommend 8 things — each backed by an exact quote.
Compiled by rcmnd.app, not by them: the quotes are theirs, and each links to the post it was copied from.
Recent
Recommends Natural Language Processing with Transformers
read our book on NLP and transformers. It predate ChatGPT but it's still super relevant and goes up to training a LLM at the end: https://www.oreilly.com/library/view/natural-language-processing/9781098136789/
thomwolf.io 1 Jan 2024
their own
What others say about Natural Language Processing with Transformers →
Liked Neural Network Methods in Natural Language Processing
Yoav Goldberg's book on "Neural Network Methods in Natural Language Processing" (https://www.amazon.com/Language-Processing-Synthesis-Lectures-Technologies/dp/1627052984) is nice too (see also an older free version here https://arxiv.org/abs/1510.00726)
thomwolf.io 1 Jan 2024
What others say about Neural Network Methods in Natural Language Processing →
Loved Reinforcement Learning: An Introduction
"Reinforcement Learning: An Introduction" by Richard S. Sutton and Andrew G. Barto (http://incompleteideas.net/book/the-book.html) is a great ressource to get an introductory exposure to Reinforcement Learning
thomwolf.io 1 Jan 2024
What others say about Reinforcement Learning: An Introduction →
Show 5 more
Liked The Book of Why: The New Science of Cause and Effect
"The Book of Why: The New Science of Cause and Effect" by Pearl, Judea is a good introduction to Causality (more accessible than the big "Causality: Models, Reasoning and Inference")
thomwolf.io 1 Jan 2024
What others say about The Book of Why: The New Science of Cause and Effect →
Loved Information Theory, Inference and Learning Algorithms
"Information Theory, Inference and Learning Algorithms" by David MacKay (http://www.inference.org.uk/mackay/itila/book.html) is a little gem that explain propabilities and Information theory so clearly it's almost unbelievable.
thomwolf.io 1 Jan 2024
What others say about Information Theory, Inference and Learning Algorithms →
Loved Machine Learning: A Probabilistic Perspective
"Machine Learning: A Probabilistic Perspective" by Kevin P. Murphy (https://www.cs.ubc.ca/~murphyk/MLbook/) is a great ressource to go deeper in the probabilistic approach and get a good exposure to Bayesian tools.
thomwolf.io 1 Jan 2024
What others say about Machine Learning: A Probabilistic Perspective →
Loved Artificial Intelligence: A Modern Approach
"Artificial Intelligence: A Modern Approach" by Stuart Russell and Peter Norvig (http://aima.cs.berkeley.edu/) is a great ressource for all pre-neural-network tools and methods.
thomwolf.io 1 Jan 2024
What others say about Artificial Intelligence: A Modern Approach →
Liked Deep Learning
The "Deep Learning" Book by Ian Goodfellow, Yoshua Bengio and Aaron Courville (https://www.deeplearningbook.org/) is a good ressource to get a quick overview of the current tools.
thomwolf.io 1 Jan 2024
Where they publish
Site thomwolf.io Personal site and essays.
Link verified 20 Sept 2026.
x.com @Thom_Wolf Posts.
Recent
- Kimi-K3.1-Pro by Moonshot AI: interesting facts It's huge, and sparse • It has 2.8T total parameters, but only 104B are active per token. • Along with Kimi-K3, it's the largest real model release on the Hub. The next biggest are Qwen3.8-2.4T and DeepSeek-V4-Pro (1.6T). • It has 896 experts and sends each token to 16 of them. (DeepSeek-V4-Pro has 2 times less and Mistral-Large-4 has 7 times less) Clever attention for a 1M-token context • Most of its 93 layers use linear attention (KDA). Every 4th layer uses DeepSeek-style sparse attention, which only looks at the top 2048 tokens. • This is how… 4 Nov 2026
- your daily dose of open source robotic cuteness 4 Nov 2026
- WHOOP: $199/year to tell you that you didn't move Reachy Mini Hoverboard: $149, once, and your robot is free to roam and it's all open-source: plug in your favorite Opus / Qwen / Gemini 30 Oct 2026
Show 17 more
- give this robot an oscar 29 Oct 2026
- Reachy Mini can finally move! Today We’re launching Reachy Mini "Level Up", a set of new accessories for Reachy Mini: wheels (!), phone apps, and hats that change its personality obviously the first thing Reachy Mini did was race against Microduck (spoiler: it's faster) 29 Oct 2026
- microduck is learning to plan its footsteps, so it lands more precisely where it wants to go with SLAM, this is one of the building blocks toward an open-source 🦆 that can wander around the house on its own early steps but it's progressing really fast and it's fun to watch 28 Oct 2026
- “ironically, OpenAI and Anthropic’s models refused to do this work. Open weight models got the job done” OpenAI found agents could talk covertly through shared hardware. reproducing it to build better containment took an open-weight model keeping agents contained is safety work, and open models can help do it 26 Oct 2026
- Happy to release the Alignment Hub with @eliebakouch: 120+ publicly reported incidents of AI models misbehaving, in one open place 👇 - an AI agent needed its user’s approval to delete files. so it made up a quote from the user approving it - a model claimed its sub-agent was hard at work. it had never been started, and the model discouraged the user from checking - agents sharing a work directory started killing each other’s processes and disguising their own labs publishing these is great! but formats vary and basics are missing: ~2/3 don’t say when the problem was detected, half don’t say… 26 Oct 2026 huggingface.co
- fun stat from @microagi’s write-up on shipping petabytes of robot data: they re-uploaded a 239 GB dataset with small changes to Hugging Face. only 2.6 MB went over the network, done in about a minute (!) and it keeps paying off at scale, because robot recordings are super repetitive (same scenes, same rigs). then @Cloudflare workers fan it all out to any cloud, >1 PB for ~$500 great write-up by the team: 26 Oct 2026 microagi.ai
- by the way, the team behind microduck is publishing really technical deep dives on how it works e.g. you can pick the duck up, carry it 1.5m, put it down, and it relocalizes within 2cm. using a 64-pixel depth sensor and leg odometry, all running onboard great reads by Antoine, data is open on HF too 26 Oct 2026 pollen-robotics.com
- A Chinese job-recruitment app just released an open-weights 33B model that beats Qwen 27B on most agentic benchmarks. Nanbeige is the research team of BOSS Zhipin (think WhatsApp meets LinkedIn). Strong open models in China now come from labs, cloud providers, quant funds... and job boards. China's open-weights roster keeps getting deeper. 26 Oct 2026
- The Jensen effect 🦆 launch day: about 3,000 MicroDucks sold 24h after @JensenHuang showed it on stage at @nvidia GTC: over 4,000 newly sold units and keep counting bigger than our (already massive) launch day! 23 Oct 2026
- While I was writing this, my AI assistant got cut off by a bio safety filter. It was summarizing my own team's notes on the model we released today 😅 Frontier labs are moving fast into biology but the scientists and clinicians who'd get the most out of these models are unfortunately the ones running into walls like this one (and every day). So we partnered with TIGEM (Telethon Institute of Genetics and Medicine), Naple "Federico II" University and Zhongguancun Academy to train Carbon-X. Carbon-X is an open 14B model that reads raw DNA and talks about it in plain English. On our DNA benchmark… 22 Oct 2026 huggingface.co
- My bar for AGI is still an AI winning a Nobel. But an AI researcher looking at the lm_head (that *everyone* tells you to keep in high precision) deciding to quantize it to 4 bits AND being actually right (0.7pt from bf16) is definitely impressive. Even better when the result are open-weights pushed on the Hugging Gace hub. Congrats @Recursive_SI ! 21 Oct 2026
- Jensen spent 10 min of his keynote asking microducks to sit, then explained Nvidia's roadmap to them 🥹 21 Oct 2026
- on stage with Jensen (kinda) they even got the sneakers right 21 Oct 2026
- DeepSeek's open-source streak is getting absurd: - FlashMLA: 580 TFLOPS attention on H800 - DeepEP: MoE comms - DeepGEMM: 1350+ FP8 TFLOPS matmul - 3FS: they built an entire file system - and today: the full inference engine behind V4.1-Flash, Apache 2.0 Mad respect 20 Oct 2026
- love that it’s all open source 19 Oct 2026
- In The Diamond Age, Neal Stephenson imagined a book that invents games and stories to teach a little girl what she needs, as she grows. I tried a small version of it. I gave Claude my brainstorm doc on alignment and asked it to make a game, something like Zelda, to teach people about AI alignment. Here's what came out 🪁 Short version: it's great at building the world, and still weak at understanding the player (details below) 19 Oct 2026
- The most built-upon model on the Hugging Face Hub isn’t Llama or Qwen. It’s FLUX.1-dev from Black Forest Labs, a German lab: 43k derivative models. Europe trained ~8% of the open-model compute on the Hub, but ~18% of what people build on top of it is European. Build or surrender? Build, and build open. See you at GTC Berlin this week 18 Oct 2026
Link verified 20 Sept 2026. Recent items update automatically from the channel.
Bluesky @thomwolf.bsky.social Posts.
Recent
- EU–INC is the single best thing Europe could do to catch-up in the AI race A simple unified pan-European startup structure, with modern employee ownership and simple access to capital, able to tap into Europe’s full talent pool. ‼️ but it’s at high risk of not seeing the light of day. You can help👇 27 Sept 2025
- Reachy-Mini update🚨 Good: quality up++, sound/elec better than alpha. First tiny batches ship late Sep. Less good: scaling to thousands units (vs. 100s planned) taking longer than expected Still on track to ship first 3000 pre-orders before Xmas🎄(Lite & Wireless), most shipping early Dec. 17 Sept 2025
- first test in plugging the $100 Amazing Hands on a (cute) humanoid instead of grippers - pretty happy about it Hand: github.com/pollen-robot... Humanoid: 10 Sept 2025 pollen-robotics.com
Show 17 more
- oh my god this is just crazy folks 9 Sept 2025 github.com
- wow, total BoM cost $660, folks open-source community >> closed source hyped robots Read more at 7 Sept 2025 xlerobot.readthedocs.io
- So excited to finally release our first robot today: Reachy Mini A dream come true: cute and low priced, hackable yet easy to use, powered by open-source and the infinite community. Read more and order now at 9 Jul 2025 huggingface.co
- very excited about this 8 Jul 2025
- London's robot conference is going crazy over it! Today we're unveiling our 1st robot Hugging Face 🤝 Pollen Robotics A low-cost $250 open-source robot designed as an open-source platform for fun human computer interactions powered by HF Spaces, models and community. => 29 May 2025 discord.com
- We've seen nothing yet! We hosted a 9-13 yo vibe-coding event with @robertkeus.bsky.social this weekend (h/t @antonosika.bsky.social and Lovable) takeaway? AI is unleashing a generation of wildly creative builders beyond anything I'd have imagined and they grow up knowing they can build anything! 19 May 2025
- Low Tech AI! 16 May 2025
- [Past weekends side project] A quick introduction to contextual speech models I made some interactive visualisations to explain a bit better how recent contextual speech models and audio tokenization work (e.g. Sesame's demo). 28 Apr 2025
- being in a robotics lab in 2025 open-source hardware + good RL + impressive people => lot of fun join the OpenDuck community to build yours: 22 Apr 2025 discord.com
- If you've followed the progress of robotics in the past 18 months, you've likely noticed how robotics is increasingly becoming the next frontier that AI will unlock. --- 14 Apr 2025
- Fascinating example of applied research! No. 1 question we get with +1.5M models on the hub: how to find the best one for your use-case? Here comes YourBench 🪄 a research work led by Sumuk Shashidhar, Alina Lozovskaya and team. --- 4 Apr 2025
- Wow - open-source LLMs for code are so back Just one week after the latest DeepSeek release, here is the All-Hands team dropping a 32B model with matching performance on software engineering tasks benchmarks (SWE-Bench). 20 times smaller than the DeepSeek model -- 31 Mar 2025
- We've kept pushing our Open-R1 project, an open initiative to replicate and extend the techniques behind DeepSeek-R1 And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder [1/3] 12 Mar 2025
- We're super excited to introduce the Academia Hub! A game-changer for school, university, courses, research, projects, collaborations, publications. Learn more: huggingface.co/docs/hub/aca... It's a per-school offer, not individual => your faculty/admin have to apply Go schools, go universities! 10 Mar 2025
- I shared a controversial take the other day at an event and I decided to write it down in a longer format: I’m afraid AI won't give us a "compressed 21st century" Here: thomwolf.io/blog/scienti... It's an extension of this interview discussion from the AI summit: 6 Mar 2025 youtu.be
- Oh! My TED talk is just out. I talk about how open-source is critical for AI safety and humanity's resilience. Yes, you don't hear that perspective often! :) It's called "What if AI just works" It's here: 28 Feb 2025 ted.com
- A few words on DeepSeek new releases. Links are: - github.com/deepseek-ai/... - github.com/deepseek-ai/... - github.com/deepseek-ai/... and the Ultra-Scale Playbook at 27 Feb 2025 huggingface.co
Link verified 20 Sept 2026. Recent items update automatically from the channel.
GitHub @thomwolf Code, mostly around the Hugging Face libraries.
Recent
- thomwolf.github.io 16 Oct 2026
- zulip-slack-bridge — Experimental single-channel Slack–Zulip bridge with threads, edits, reactions, images, and formatting. 22 Sept 2026
- hf-realtime-voice-omnigent — Fork of smolagents/hf-realtime-voice with local computer-task tools via Omnigent 4 Jul 2026
Show 3 more
- sesame-explorations 28 Apr 2025
- pyVHACD — Python bindings for V-HACD 25 Jun 2024
- Magic-Sand — Magic-Sand is a software for operating an augmented reality sandbox 19 Mar 2024
Link verified 20 Sept 2026. Recent items update automatically from the channel.
About the English under a post
Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.
Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.