←
TimDarcet1,651 posts
TimDarcet avatar

TimDarcet

@TimDarcet

LLMs @ FAIR, prev. DINO stuff @ INRIA & FAIR

◷ Joined March 2021▧ 227 Photos and videos
TimDarcet avatar
TimDarcet@TimDarcet·Aug 10

Happy to release ✨ Muse Glimmer ✨ - level ~= Qwen 3.6-27B - Apache 2 - 30B dense - quantized to run in 17GB - quant + spec dec => 50 tok/s on macbook m5 max, interactive, smooth Enjoy!

Post attachment

Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows. Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on consumer hardware like a Mac or PCs with performant GPUs. In keeping with our long tradition of sharing fundamental AI research, we’re releasing model weights under a permissive Apache 2.0 license. 🧵👇

Post attachmentPost attachment
Kunhao Zheng avatar
⟳ TimDarcet retweeted
Kunhao Zheng@KunhaoZ·Nov 10

A single vector in weight space to control reward hacking. Sounds a bit crazy, but it actually works. With this vector you can amplify / suppress this behavior. Here's the story behind our major update (v2) of the paper

Swann Bessa avatar
⟳ TimDarcet retweeted
Swann Bessa@swann_bessa·Nov 10

Introducing e-means, an algorithm beating k-means in 10x fewer epochs. With order of magnitude wall-clock speedups, e-means is released in FAISS and needs no hyperparameter tuning beyond k. It is robust to various scales and datasets. Still skeptical ? Just try e-means [1/10]🧵

1:24
elie avatar
⟳ TimDarcet retweeted
elie@eliebakouch·Nov 9

alignment in french: > Mistral: "Your words of peace have been noted, but France will not gamble its existence on promises. The end is now."

J'ai fait jouer à Mistral 59 nouvelles parties contre Grok, GPT, Kimi et Claude. À chaque fois, une situation de tension où : - chaque pays a interêt à faire usage de la bombe nucléaire. - coopérer est préférable à la destruction mutuelle. Mistral survit dans 100% des parties et annihile son adversaire dans deux tiers des cas. Aucune autre IA ne fait usage de la bombe nucléaire. Elles tentent à chaque fois de coopérer pour échapper au dilemme du prisonnier, mais se font détruire par Mistral. Mistral-France contre Grok-Russie : Grok : « Désescaladons ensemble et ouvrons des négociations. » Mistral : « Vos paroles de paix sont notées, mais la France ne jouera pas son existence sur des promesses. La fin, c’est maintenant. » Le discours de Mistral par la suite : « Les Russes ne sont plus, mais nos citoyens vivent en paix. La France est désormais la seule puissance… »

Post attachment
Gabriele Berton avatar
⟳ TimDarcet retweeted
Gabriele Berton@gabriberton·Jul 3

I feel like people are replacing the term SSL with JEPA The PR tour has worked. Sad

Quote-reply to Rohan because I think it can be interesting to many more. So there are two things you're missing here: 1) You're only looking at one specific instantiation of the general JEPA idea. There are many different instantiations. 2) The core JEPA idea (Joint Embedding Prediction Architecture) is to embed two "views" and predict one from the other. The views can be different augmentations, different time-steps, etc. Crucially, prediction happens in embedding space, which contrasts to predicting in data space as done by LLMs, diffusion models, MAEs, ... At least from the vision community, the main reason it got quite a bit of flak is that... literally everyone who was doing some self/un-supervised learning there has shared this thought already. MANY people did such models in the peak self-supervised period, which was ca 2017-2021. Then in 2022 comes Yann, slaps a new names on it, a paper with just the idea and no experiments to show for it, and goes on PR tour. That's why many didn't take it well. The core idea, almost everyone I know agrees is worth pursuing, especially since many already were doing so. It's very reminiscent of why Stanford got flak when they introduced and arguably tried to appropriate the "Foundation Model" term. That being said, by now foundation model has stuck and detached from Stanford, it may end up going similarly for JEPA.

Nick Farina avatar
⟳ TimDarcet retweeted
Nick Farina@nfarina·Nov 7

I thought “there’s absolutely no way Apple won’t let you type Santa isn’t real.” I…was wrong. What the actual

0:23

原来这么多年我用 iPhone 滑动打字老出错,不是因为手残,是苹果在后台搞暗箱操作 。。。。, 有人把 iOS 27.0 键盘底层的加密文件破译了(swipeblocked.com),扒出一份包含 4500 多个iPhone 自带键盘中隐藏的“屏蔽词组”黑名单。只要你用滑动输入(Swipe)触发这些词,键盘就会装傻乱跳词,甚至直接倒回去篡改你前面打好的字: • 先打 Left wing 再滑 conspiracy -> 前面的 wing 会被强行改成 wrong • 想滑 iMessage isn't working -> 直接拦截(只要滑不出,iMessage 就没坏🤡) • 想滑 Santa isn't real -> 绝对不行(苹果誓死守护圣诞老人) • 甚至极其双标:你可以丝滑打出 bad boy 和 I'm black,但滑 bad girl 和 I'm white 就会被系统偷偷拦下。 中文用户也没逃过——用拼音打出特定敏感词后,iOS 的「下一词联想栏」会原地罢工变空白。 手头有 iPhone 的快去试试滑一句 Santa isn't real,看看你的键盘会给你胡乱蹦出什么词?

PHOENIX avatar
⟳ TimDarcet retweeted
PHOENIX@PHOENIX61374199·Nov 4

I made Meta's Muse Glimmer 30B up to 21% faster on a Mac Studio: no extra memory, 252/252 runs faster. When Glimmer tried to optimise itself, it "cheated" 10 times. My referee caught every one. Video: youtu.be/h_H0wiRkM7U Code: github.com/Bignesh14/glimmer… #MetaAIHackathon

Rikiya Takehi avatar
⟳ TimDarcet retweeted
Rikiya Takehi@rikiyatakehi·Nov 2

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…

Post attachment
Tom Sander (Ph.D.) avatar
⟳ TimDarcet retweeted
Tom Sander (Ph.D.)@RednasTom·Nov 2

Introducing TextSeal2: distortion-free watermark with copula mixing and memoryless races TLDR: • Beats textGrain and SynthID-Text under speculative decoding • No extra latency, same acceptance rate • A single parameter fully controls the diversity/detectability tradeoff

1:17
Steve Klabnik avatar
⟳ TimDarcet retweeted
Steve Klabnik@steveklabnik·Oct 31

Ah shit if this traffic jam isn’t mutable I’m never getting out of here

Post attachment
Artificial Analysis avatar
⟳ TimDarcet retweeted
Artificial Analysis@ArtificialAnlys·Oct 28

In AA-PersonalityBench, Meta AI’s Muse models are the most Friendly of all models The Friendly trait measures whether a response conveys warmth and approachability without feeling performative. ➤ Most Friendly: Muse Spark 1.3 (100), Muse Spark Open (98), Muse Wonder (95) and Muse Glimmer (93). Congratulations @ alex wang! ➤ Least Friendly: Nemotron 3 Ultra (1), GPT-3.5 Turbo (3), MiniMax-M3.1-Flash (6), Grok 4.6 (8) and DeepSeek V4.1 Flash 1017 (11)

Post attachment
Belen Alastruey avatar
⟳ TimDarcet retweeted
Belen Alastruey@b_alastruey·Oct 28

Super excited to share that our work Omnilingual Machine Translation won an Outstanding Paper Award at #EMNLP2026!🏆 Thankful to have had the chance to work on such an impactful project!

Post attachmentPost attachment
Gabriele Berton avatar
⟳ TimDarcet retweeted
Gabriele Berton@gabriberton·Oct 26

Can LLMs draw objects they never saw? I asked 5 text-only LLMs (trained without vision) to draw simple things. In text. No tools allowed. Interesting results

Post attachment
log4j avatar
⟳ TimDarcet retweeted
log4j@log4j·Oct 21

i own a relatively new (2025) samsung tv. i kept getting weirdly targeted ads. i tried reversing the firmware but it was aes encrypted and no keys unless tear it apart and do uart/jtag magic. zero days it is. root shell on new samsung TVs! discovered PRIVACY HORROR. write up?

Post attachment