clem &#x1f917;(@ClementDelangue)

Shared my first trace from @NanoClaw_AI to @huggingface yesterday. Very cool! By default, all agen...

clem 🤗 (@ClementDelangue) — Thu, 04 Jun 2026 14:31:17 GMT

Shared my first trace from @NanoClaw_AI to @huggingface yesterday. Very cool! 

By default, all agents should store their traces on HF (in private) so that you can keep a history of them, analyze them,... & share them and post-train better models, harnesses and more. Excited about this direction!

💬25🔄15❤️90👀13781📊38

At this point, I suspect you could put endpoints named 0pus 4.8 & GPT 5.S in your apps powered b...

clem 🤗 (@ClementDelangue) — Wed, 03 Jun 2026 21:19:17 GMT

At this point, I suspect you could put endpoints named 0pus 4.8 & GPT 5.S in your apps powered by open-source models and it would get massive usage without people complaining. The power of "frontier" marketing!

💬47🔄7❤️273👀25299📊66

Routing and post-training open-source models won't only give you more accurate systems but also mean...

clem 🤗 (@ClementDelangue) — Wed, 03 Jun 2026 19:03:03 GMT

Routing and post-training open-source models won't only give you more accurate systems but also meaningfully faster and cheaper systems as most companies are currently learning (in addition to giving you more control and privacy).

The idea that a "frontier" model (by frontier we mean is slightly more accurate on a few very limited benchmarks) will be better for all domains, all tasks, all setups just doesn't hold up! It's marketing for making you pay more!

Harvey@harvey

We partnered with @FireworksAI_HQ to train open-source models for legal. Here's what we found:

1) Hybrid legal agents can beat frontier models on quality and cost by routing selectively to a frontier advisor.

We tested a hybrid setup where GLM 5.1 served as the primary worker, routing tasks to Opus 4.7 as an advisor when needed.

GLM invoked Opus sparingly, just 0.83 times per task on average.

The hybrid setup beat Opus on both quality and cost: 18% all-pass vs 14%, at $368 vs $954 across the same 100 tasks.

2) Post-training can push open models to frontier-level legal performance.

On a 100-task slice of our Legal Agent Benchmark (LAB), SFT moved Kimi 2.6's all-pass rate from 11% to 15%, beating Opus' 14%.

But the cost gap was even more striking: $84 vs $954 across the same 100 tasks, or ~11x cheaper.

We're excited to continue working with @FireworksAI_HQ on the next generation of open-source legal agents.

💬27🔄28❤️270👀26467📊53

Arcee needs more attention that it gets! There aren't a lot of great American open-source AI model c...

clem 🤗 (@ClementDelangue) — Tue, 02 Jun 2026 20:29:42 GMT

Arcee needs more attention that it gets! There aren't a lot of great American open-source AI model companies and they're one of them! huggingface.co/arcee-ai

💬12🔄28❤️175👀15453📊35

Automatic behind the scene routing in user interfaces (instead of model picker) will redistribute va...

clem 🤗 (@ClementDelangue) — Tue, 02 Jun 2026 18:02:14 GMT

Automatic behind the scene routing in user interfaces (instead of model picker) will redistribute value capture and usage towards many more models than just frontier ones (especially towards open-source/smaller/cheaper ones). Because it removes the cognitive load for the final user to have to switch models (which is too high leading to defaulting to frontier for all calls with model picker) cc @antonosika@bgurley

Factory@FactoryAI

Introducing model routing to Factory.

Factory Router picks the right model for every task, automatically.

Maintain frontier performance while cutting costs by 25%.

Your browser does not support the video tag.

💬15🔄18❤️159👀23155📊29

It's cool to see that the https://t.co/SdGWBBVrqG has become progressively a major source of learnin...

clem 🤗 (@ClementDelangue) — Mon, 01 Jun 2026 21:28:19 GMT

It's cool to see that the hf.co/blog has become progressively a major source of learning and news about AI for the community.

Just look at the recent content created & major announcements & tutorial from @openclaw, @nvidia, @jetbrains, @IBMResearch,...

What is the next blogpost you're going to create on HF? AI need to be more public, open and collaborative and we need everybody's help to make it so!

💬24🔄19❤️141👀19571📊42

So much great work lately from Nvidia, the "King of American Open-source AI"! - Crossed 1,000 total...

clem 🤗 (@ClementDelangue) — Mon, 01 Jun 2026 16:36:35 GMT

So much great work lately from Nvidia, the "King of American Open-source AI"!

- Crossed 1,000 total public repositories on @huggingface (820 models, 249 datasets & 57 spaces) & almost 60,000 followers
- Current #1 trending model on HF with LocateAnything and #5 trending with PiD
- Announced that they're adopting the @linuxfoundation OpenMDW framework
- Released Cosmos 3, Omnimodal World Models for Physical AI & Alphamayo 2 Super, an open model for autonomous driving
- Announced the release soon of Nemotron 3 & work on Nemotron 4

Thank you @nvidia for all the work you're doing for the ecosystem and open-source AI. Can't wait for the next few months!

💬24🔄58❤️434👀102938📊85

Feels quite magical to be able to clone a 68 TB dataset to my private HF training bucket while I onl...

clem 🤗 (@ClementDelangue) — Mon, 01 Jun 2026 12:40:57 GMT

Feels quite magical to be able to clone a 68 TB dataset to my private HF training bucket while I only have a 4TB local disk, all of that in less than a minute thanks to HF infra optimizations & xet dedup!

💬14🔄9❤️133👀13674📊28

We need more coding and agent traces public sharing to build datasets and better open source models!...

clem 🤗 (@ClementDelangue) — Sun, 31 May 2026 21:23:42 GMT

We need more coding and agent traces public sharing to build datasets and better open source models! Lots of people contributing already, you should share yours too! huggingface.co/datasets?searc…

Simon Willison@simonw

I'm really upset about this: OpenAI's Codex Desktop had a "Copy as Markdown" option for exporting full chat transcripts, but the feature vanished in an update a couple of days ago

Genuinely my single favorite feature of Codex compared to Claude Code
 github.com/openai/codex/i…

💬21🔄23❤️192👀44671📊38

AI safety can't happen behind closed doors! Super cool to see that the @AISecurityInst is releasing ...

clem 🤗 (@ClementDelangue) — Sat, 30 May 2026 15:43:45 GMT

AI safety can't happen behind closed doors! Super cool to see that the @AISecurityInst is releasing its evals, datasets, and models in the open on @huggingface, so researchers everywhere can scrutinize, reproduce, and build on them: huggingface.co/ai-safety-inst…

💬21🔄45❤️139👀25366📊48

Agree! was talking about this with @havoyan just a few days ago. That's also the reason why so much ...

clem 🤗 (@ClementDelangue) — Fri, 29 May 2026 16:01:56 GMT

Agree! was talking about this with @havoyan just a few days ago. That's also the reason why so much of the value has been accruing to the frontier models in my opinion (cc @GavinSBaker) because if you don't have the mental load to pick/change models, you'll just default to the one that has been marketed as "frontier" even if it's not the best trade-off for the majority of things you're doing with it (ex you can get cheaper more efficient models on telling you the weather, classifying, question answering,...)

Anton Osika@antonosika

I don’t think most people actually want to choose between AI models.

Every week there’s new consensus on what’s best, but most people just want great results, and they don’t care which model powers that.

At Lovable, we spend a lot of time evaluating models internally so our users don’t have to. The goal is simple: automatically use the best model for the task so people can stay focused on building their business.

Don't underestimate how important it is to remove mental load for your customers at every step of the way.

💬8🔄1❤️26👀6976📊8

You pick AI models based on their model cards, why not humans haha. Well done Noah! https://t.co/3PQ...

clem 🤗 (@ClementDelangue) — Fri, 29 May 2026 15:43:47 GMT

You pick AI models based on their model cards, why not humans haha. Well done Noah! huggingface.co/noahmclaughlin…

💬8🔄5❤️39👀10340📊12

@kevincodex much cheaper than s3/r2 thanks to xet: https://t.co/E6lTU1CRyV

clem 🤗 (@ClementDelangue) — Fri, 29 May 2026 15:14:38 GMT

@kevincodex much cheaper than s3/r2 thanks to xet: hf.co/storage

💬3🔄5❤️38👀5110📊8

Most people know Hugging Face from its public models and datasets but few realize that 50% of the mo...

clem 🤗 (@ClementDelangue) — Fri, 29 May 2026 15:10:54 GMT

Most people know Hugging Face from its public models and datasets but few realize that 50% of the models and datasets stored on HF are private. 

This number has been increasing with buckets (our S3 alternative for AI) and enable companies to build AI more efficiently and collaboratively within their organizations, even when they don't share publicly!

Excited to see more of that in the coming months as more companies start building AI themselves instead of outsourcing to APIs!

💬22🔄31❤️265👀26059📊57

Most people training agentic LLMs with RL right now have a silently broken training loop and have no...

clem 🤗 (@ClementDelangue) — Fri, 29 May 2026 01:44:09 GMT

Most people training agentic LLMs with RL right now have a silently broken training loop and have no idea.

Here's the trap: single-turn RL works beautifully. Clean curves, sane rewards, everything converges. Then you add tools so the model can act mid-rollout, and things get weird. Loss spikes for no reason. Eventually a shape-mismatch error.

The culprit: every time you parse the model's output to detect a tool call, then re-tokenize the updated conversation for the next turn, you're rolling the dice. Usually the round-trip gives back the same tokens. Sometimes it doesn't and your gradient lands on a sequence the model never actually sampled. No crash. Just quietly wrong math and a useless gradient signal.

The fix is one rule: never re-encode tokens you've decoded. Keep the sampled tokens in one buffer, never re-render them, and both failure modes disappear. That's Token-In, Token-Out done right.

Our team just published a beautiful deep-dive on exactly this, including an audit across the major open-weights model families showing most chat templates already support it. Required reading if you're doing multi-turn RL 🤗🔥qgallouedec-tito.hf.spacejM

💬49🔄139❤️1112👀1009960📊332

The HF science team just made async RL weight sync ~100x cheaper on bandwidth, and you don't need a ...

clem 🤗 (@ClementDelangue) — Thu, 28 May 2026 13:23:56 GMT

The HF science team just made async RL weight sync ~100x cheaper on bandwidth, and you don't need a shared cluster anymore.

The problem: every RL step, the trainer typically has to sync fresh weights to the inference engine. for a 7B in bf16 that's ~14GB. for a frontier 1T fp8 checkpoint, that's ~1TB; in bf16 it would be ~2TB. per sync.

The insight: between two RL steps, ~99% of bf16 weights are bit-identical. at RL learning rates, the optimizer is whispering and bf16 literally cannot hear most of it. the stored bf16 bits don't change.

What they shipped in TRL: only the changed elements get encoded as a sparse safetensors file, dropped into a Hugging Face Bucket, and fetched by vLLM. on Qwen3-0.6B, per-step payload goes from 1.2 GB to 20 to 35 MB. This is exactly what we built Buckets for: S3-like object storage on the Hub, Xet-backed (so even full snapshots only transfer the changed chunks).

The cherry on top: we ran a FULL disaggregated training where:
- the trainer lived on one box
- vLLM ran inside a Hugging Face Space
- the Wordle environment ran in another Space
- weights flowed through one Hub bucket

no shared cluster. no RDMA. no VPN. no NCCL across clouds. just HTTPS and a bucket.

one GPU + a Hugging Face account is now enough to do real disaggregated RL. multi-replica inference fleets across regions become a small devops exercise, not a research project.

Full write-up: huggingface.co/blog/delta-wei…

Open source RL keeps eating the moat!

💬28🔄68❤️592👀58412📊143

@josefchen Nice! Could you add the model and dataset to HF?

clem 🤗 (@ClementDelangue) — Wed, 27 May 2026 13:36:29 GMT

@josefchen Nice! Could you add the model and dataset to HF?

💬1🔄0❤️94👀61967📊9

Today, I'm working on trying to better understand how coding assistants mention HF's products. Tak...

clem 🤗 (@ClementDelangue) — Mon, 25 May 2026 19:26:19 GMT

Today, I'm working on trying to better understand how coding assistants mention HF's products. 

Taking a simple approach of running tons of queries and analyzing the answers with @DAKlingbeil's submarine.ai (ex: huggingface.co/datasets/clem/…) 

Are there better/different approaches? What has been working for you?

💬18🔄2❤️58👀7139📊23

Is this advertising?

clem 🤗 (@ClementDelangue) — Mon, 25 May 2026 18:58:55 GMT

Is this advertising?

💬10🔄2❤️90👀19258📊19

llama.cpp with MTP support makes local models fast enough to use as daily drivers 🚀 Qwen3.6-27B d...

clem 🤗 (@ClementDelangue) — Sun, 24 May 2026 22:12:01 GMT

llama.cpp with MTP support makes local models fast enough to use as daily drivers 🚀 

Qwen3.6-27B dense generation below on A10G:  From 25 tok/st to 45 tok/s (+78%)!

Your browser does not support the video tag.

💬82🔄118❤️1339👀137528📊349

300,000 AI builders filled their hardware profile on @huggingface and we're sharing the results: htt...

clem 🤗 (@ClementDelangue) — Sun, 24 May 2026 16:56:23 GMT

300,000 AI builders filled their hardware profile on @huggingface and we're sharing the results: hf.co/hardware. 

Excited to see how it evolves in the coming months especially with the explosion of local AI!

💬38🔄39❤️239👀39751📊76

Yesterday, for the winning French cup final, there was 50,000 Lens fans at the stade de France + 40,...

clem 🤗 (@ClementDelangue) — Sat, 23 May 2026 10:44:47 GMT

Yesterday, for the winning French cup final, there was 50,000 Lens fans at the stade de France + 40,000 at the home stadium + 10,000 in the fan zones = 100,000 total. The city only has 33,000 inhabitants 🤯🤯🤯 

Is there any equivalent in the worl#rclensens

💬21🔄25❤️402👀43147📊59

Yes! but why the evil HF logo haha https://t.co/wIlEq66HrG

clem 🤗 (@ClementDelangue) — Fri, 22 May 2026 13:51:39 GMT

Yes! but why the evil HF logo haha spectrum.ieee.org/open-source-ro…

💬11🔄8❤️121👀29914📊24

reachy mini passed out on @lalopenguin's live cooking show (wait for it)

clem 🤗 (@ClementDelangue) — Fri, 22 May 2026 12:52:06 GMT

reachy mini passed out on @lalopenguin's live cooking show (wait for it)

Your browser does not support the video tag.

💬6🔄6❤️51👀26421📊13

Great to see @CommonCrawl using and recommending @huggingface Buckets for large constantly evolving ...

clem 🤗 (@ClementDelangue) — Fri, 22 May 2026 12:31:51 GMT

Great to see @CommonCrawl using and recommending @huggingface Buckets for large constantly evolving training datasets! If you have private models or datasets, try it and let us know what you think about it! huggingface.co/storage

💬6🔄6❤️58👀12530📊11

Ready for the French cup final today! #allezlens

clem 🤗 (@ClementDelangue) — Fri, 22 May 2026 12:14:52 GMT

Ready for the French cup final today! #allezlens

💬10🔄0❤️40👀5563📊11

I'm excited about the new @amd Ryzen AI Halo because we need more local hardware for AI builders! T...

clem 🤗 (@ClementDelangue) — Thu, 21 May 2026 21:30:06 GMT

I'm excited about the new @amd Ryzen AI Halo because we need more local hardware for AI builders!

There's something fun and exciting about building on your own machines rather than sending to the cloud! Should we do our own @huggingface hardware for AI builders at some point?

Your browser does not support the video tag.

💬25🔄30❤️273👀41854📊56

Looks like I was just two years early haha

clem 🤗 (@ClementDelangue) — Thu, 21 May 2026 21:06:18 GMT

Looks like I was *just* two years early haha

💬11🔄4❤️62👀5669📊17

the reason is marketing!

clem 🤗 (@ClementDelangue) — Thu, 21 May 2026 19:41:45 GMT

the reason is marketing!

Patrick OShaughnessy@patrick_oshag

Interesting that in three of my recent conversations with Krishna at Anthropic, @dylan522p, and @GavinSBaker, each said that frontier tokens are capturing the majority of the economic value.

Gavin: "An overwhelming amount of the economic returns to AI at the model layer have been at the frontier. That's surprising to me, and I think it's been surprising to a lot of people. This is one of the most important questions to be answered, and you need to have a hypothesis on it as an investor."

Krishna: "We think the returns to frontier intelligence are extremely high. Customers invest really heavily in more tokens with the newer models. The ones at the frontier clearly are capturing this economic value, driving meaningful ROI for customers. The returns to frontier intelligence are not slowing down."

Dylan: "No one gives a crap about GPT-4 class models. They want the frontier because the frontier lets them create the economically valuable things."

Your browser does not support the video tag.

💬7🔄2❤️25👀6963📊7

Cohere is on such a great open-source trajectory lately. Beautiful Apache 2.0 model! https://t.co/Be...

clem 🤗 (@ClementDelangue) — Wed, 20 May 2026 19:22:00 GMT

Cohere is on such a great open-source trajectory lately. Beautiful Apache 2.0 model! huggingface.co/CohereLabs/com…

Nick Frosst@nickfrosst

Command A+ from @cohere  is out now :) its our best model yet and its open source apache 2.0

Your browser does not support the video tag.

💬6🔄34❤️316👀39094📊42

clem 🤗 (@ClementDelangue) — Wed, 20 May 2026 19:06:51 GMT

💬6🔄6❤️46👀10821📊10

The future of biology shouldn’t stay behind black-box APIs. Especially when it touches personal heal...

clem 🤗 (@ClementDelangue) — Wed, 20 May 2026 12:10:50 GMT

The future of biology shouldn’t stay behind black-box APIs. Especially when it touches personal health.

Whether you’re @bryan_johnson measuring every biomarker, or @sytses openly sharing and analyzing his own immune-genetics data, you need open, local, transparent AI.

@huggingface wasn’t created to be a biology company. It’s not the most obvious focus for us. But it feels too important not to do something.

That’s why we built and released Carbon 🧬: a frontier DNA base model with open weights, training code and data pipeline, designed to be fine-tuned or continually pretrained for downstream biological tasks.

Carbon is 275x faster than the next best model at its size. Fast enough to run locally on your laptop. Powerful enough to process a whole human genome on a single GPU in less than 2 days.

The technical unlock: a DNA-native tokenizer that splits sequences into 6-base chunks for efficiency, while preserving single-base resolution during training and inference. More people able to inspect, run, fine-tune, improve and build on top of the models shaping biology.

Open weights:huggingface.co/collections/Hu…q
Dataset:huggingface.co/datasets/Huggi…P 
Demo:huggingface.co/spaces/Hugging…b

Let's go open AI biology!

💬18🔄50❤️241👀24171📊61

I remember when people were saying "It's useless to open-source big models because nobody will be ab...

clem 🤗 (@ClementDelangue) — Tue, 19 May 2026 23:10:26 GMT

I remember when people were saying "It's useless to open-source big models because nobody will be able to run them fast"....

Cerebras@cerebras

Cerebras is now running Kimi K2.6 – a trillion parameter model – in enterprise trials. 

At ~1,000 tokens/s, this is the fastest frontier model performance ever measured by Artificial Analysis @ArtificialAnlys.

💬59🔄132❤️2136👀320822📊256

New dances added to the reachy mini desktop app. Try them and post a video below?

clem 🤗 (@ClementDelangue) — Tue, 19 May 2026 18:44:22 GMT

New dances added to the reachy mini desktop app. Try them and post a video below?

💬10🔄7❤️50👀4970📊16

what? we need more support for local options from inference engines, not less!

clem 🤗 (@ClementDelangue) — Tue, 19 May 2026 16:52:57 GMT

what? we need more support for local options from inference engines, not less!

💬18🔄7❤️159👀21118📊33

.@karpathy at @AnthropicAI = more open-source from them? They're already contributing datasets: http...

clem 🤗 (@ClementDelangue) — Tue, 19 May 2026 16:29:59 GMT

.@karpathy at @AnthropicAI = more open-source from them? They're already contributing datasets: huggingface.co/Anthropic but would be super cool to see more (massive potential for impact obviously!)

💬16🔄23❤️243👀14924📊44

My number of followers on X has been growing much faster than on other social media these days. Grea...

clem 🤗 (@ClementDelangue) — Mon, 18 May 2026 19:17:35 GMT

My number of followers on X has been growing much faster than on other social media these days. Great job @nikitabier and team!

💬16🔄0❤️106👀6143📊22

I believe on-prem and local AI - based on @huggingface open-source models - will be an important ans...

clem 🤗 (@ClementDelangue) — Mon, 18 May 2026 18:18:44 GMT

I believe on-prem and local AI - based on @huggingface open-source models - will be an important answer to the GPU shortages this year (because they are cheaper, faster, safer than cloud APIs)!

Great collaboration between @huggingface & @MichaelDell@Dell to make this a reality for enterprise today. Announced at the main keynote of Dell Technologies World.

Your browser does not support the video tag.

💬53🔄57❤️446👀74795📊121

Reachy mini rap battle at @aiDotEngineer Singapore!

clem 🤗 (@ClementDelangue) — Mon, 18 May 2026 17:37:21 GMT

Reachy mini rap battle at @aiDotEngineer Singapore!

Your browser does not support the video tag.

💬8🔄13❤️105👀25738📊20

Very cool to see Cursor doubling down on training great models. In my opinion, ultimately all seriou...

clem 🤗 (@ClementDelangue) — Mon, 18 May 2026 17:27:17 GMT

Very cool to see Cursor doubling down on training great models. In my opinion, ultimately all serious companies in AI will want to train models themselves, based on open-source instead of outsourcing AI to others via APIs!

Cursor@cursor_ai

Introducing Composer 2.5, our most powerful model yet.

It's more intelligent, better at sustained work on long-running tasks, and more reliable at following complex instructions.

For the next week, we’re doubling the included usage of the model.

💬18🔄31❤️238👀27101📊33

Paper of the day! https://t.co/6LSlvyBYKh

clem 🤗 (@ClementDelangue) — Fri, 15 May 2026 17:03:37 GMT

Paper of the day! huggingface.co/papers/2605.13…

Ning Ding@stingning

We’re releasing a 30B-A3B reasoning model that reaches gold-medal level across both physics and math Olympiad evaluations: IPhO directly, and IMO/USAMO with test-time self-verification and refinement.

A simple, unified scaling recipe for proof search.

huggingface.co/papers/2605.13…

💬6🔄45❤️289👀61675📊56

AI teams shouldn’t have to choose between expensive object storage and painful git workflows. @hugg...

clem 🤗 (@ClementDelangue) — Fri, 15 May 2026 15:07:46 GMT

AI teams shouldn’t have to choose between expensive object storage and painful git workflows.

@huggingface Storage is built for model weights, datasets, checkpoints and artifacts:
- simple per-TB pricing
- built-in CDN
- Xet deduplication
- private by default when needed

Store your AI data where your AI work already happens: huggingface.co/storage

💬10🔄8❤️94👀43812📊18

Are scaling laws finally working for time series foundation models? Today, @datadoghq is releasing ...

clem 🤗 (@ClementDelangue) — Thu, 14 May 2026 18:24:52 GMT

Are scaling laws finally working for time series foundation models?

Today, @datadoghq is releasing Toto 2.0 weights in Apache 2.0 on @huggingface. It's a family of open-weights TSFMs from 4M to 2.5B parameters, where every size beats the last from a single hyperparameter config. First across the leading benchmarks: BOOM, GIFT-Eval, and TIME.

Most TSFM families ship multiple sizes that all perform roughly the same. This one doesn't.

Why it matters: scaling laws gave language and vision a predictable relationship between compute, data, parameters, and downstream performance. Time series hasn't had that curve until now. Once you have it, you can scale data and compute with confidence, and start asking which new capabilities emerge at the next order of magnitude.

2.5B open-source weights: huggingface.co/Datadog/Toto-2…
4M open-source weights: huggingface.co/Datadog/Toto-2…

Blogpost: datadoghq.com/blog/ai/toto-2…

💬9🔄30❤️200👀24265📊45

are you ready for portable reachy mini?

clem 🤗 (@ClementDelangue) — Thu, 14 May 2026 14:29:08 GMT

are you ready for portable reachy mini?

Your browser does not support the video tag.

💬18🔄14❤️163👀25218📊32

Did @Jason just pull up in Miami?

clem 🤗 (@ClementDelangue) — Thu, 14 May 2026 13:39:55 GMT

Did @Jason just pull up in Miami?

💬9🔄1❤️194👀133399📊26

Good catch @theojaffee 😂

clem 🤗 (@ClementDelangue) — Wed, 13 May 2026 23:45:59 GMT

Good catch @theojaffee 😂

Your browser does not support the video tag.

MTS@MTSlive

We asked the CEO of HuggingFace @ClementDelangue what the risks of releasing powerful open source models are.

He says restricting AI creates more risk than openness.

"Six, seven years ago, at the time it was GPT-2, and there was already a lot of people saying that it was too dangerous to release in open source."

"Mythos, when it was announced was crazy dangerous... In a few weeks or a few months, everyone is gonna be using Mythos, and not destroy the world as a result."

"For cybersecurity, the biggest risk is that a few players have capabilities that other people don't have... If you make it more open, it's usually easier for defenders to react and make the whole system safer."

"The idea of restricting a technology like AI based on risks is like saying, 'Some people can punch other people, so let's tie down everybody's hands.'"

"Otherwise you slow down progress, you create massive gaps in terms of controls, in terms of capabilities, and you create actually additional risks."

Your browser does not support the video tag.

💬6🔄4❤️51👀11371📊9

Is that more or less accurate? Bullish for local AI as a solution to the energy & cloud shortage...

clem 🤗 (@ClementDelangue) — Wed, 13 May 2026 15:07:39 GMT

Is that more or less accurate? Bullish for local AI as a solution to the energy & cloud shortages!

💬13🔄12❤️63👀5774📊22

As President Trump meets President Xi this week, a call to the American AI community: If your start...

clem 🤗 (@ClementDelangue) — Wed, 13 May 2026 11:02:13 GMT

As President Trump meets President Xi this week, a call to the American AI community:

If your startup, lab, non-profit or company benefits from open international AI - especially Chinese (Deepseek, Qwen, Kimi, GLM,…), please share!

Open source is the most important driver of competition, jobs and wealth creation in AI today. Let’s support and promote it at critical times like this week!

💬33🔄73❤️545👀78638📊98