François Porcher (@foporcher) header
François Porcher (@foporcher) profile picture

PhDing @AmiLabs and @Inria | World Modeling, Planning, Robotics | prev. SAM3 @Meta | @UCBerkeley

Paris, France
Joined November 2023
Photo by François Porcher (@foporcher)Photo by François Porcher (@foporcher)Photo by François Porcher (@foporcher)Photo by François Porcher (@foporcher)Photo by François Porcher (@foporcher)Photo by François Porcher (@foporcher)Photo by François Porcher (@foporcher)
1/ My first PhD paper is out! 🎓 Title: Flow Matching in Feature Space for Stochastic World Modeling tldr: we build stochastic world models directly in high-dimensional DINOv3 feature space, instead of relying on low-dimensional VAE latents.
24
89
6
849
104,230
François Porcher retweeted
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
37
57
11
457
24,630
François Porcher retweeted
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]🧵
150
1,223
257
10,224
892,234
🚀 I'm transferring my PhD to @amilabs from @AIatMeta ! A PhD is the time to take bets, and to back the paradigm shifts you believe in. For the second half of mine, I’ll be pushing on World Models, Robotics and Planning 🌍🤖 Grateful for my time at FAIR: SAM3 with @alcinos26 , FlowWM (arxiv.org/abs/2606.29059), and more soon with the RL Team led by @alelazaric 👀 Excited to work alongside @pascalefung @garridoq_ @AdrienBardes , and thanks to my @Inria supervisors Shizhe Chen and @inthebrownbag for supporting this move 🙏🏻 Push push! 💪🏻
9
4
1
213
13,519
Neat paper from @mathuvu_ and @byoubii ! They show that standard Chinchilla scaling law misses an interaction between model size and data. Their fix is simple: add one coupling exponent, which gives more accurate predictions of the compute-optimal model/data allocation.
Chinchilla writes loss as a sum of two independent terms: model size and data never interact. Kaplan's original law coupled them. But interaction has an impact, dropping it bends predictions exactly where you extrapolate. Skaling, our new law, puts it back with one exponent. 1/8
7
1,072
François Porcher retweeted
l’ingénieur a 17h qui fait un last try avec des prompts « believe in yourself » et « keep going » et réalise une avancée scientifique 🤣
We asked an unreleased research version of Claude to take a stab at the Riemann hypothesis. It didn’t solve it, but it did make strides on a related problem: it increased the lower bound for the fraction of zeros of the Riemann zeta function that satisfy the hypothesis from 41.6% to 67.2%. anthropic.com/research/riema…
5
44
3
1,407
129,681
Great work from @KrunoLehman and @webalorn ! WaiT introduces frequency-aware flow matching: generate coarse structure first, then progressively denoises high-frequency details with wavelet-specific schedules. A simple change which improves image and video generation
WaiT for the Signal: Simple Frequency-Aware Flow-Matching. We show how to natively incorporate a fundamental property of images directly into diffusion models; setting a new pixel-space SOTA on ImageNet, while reducing compute. 📄 arxiv.org/abs/2607.28760 Full breakdown below👇
1
5
1,398
FAIR rising-star 💫 PhD student @PierreChambon6 strikes again: his new paper shows how to teach LLMs to generate code that is not only correct, but genuinely fast, by turning noisy execution times into a reliable reinforcement-learning signal.
🧵How to combine code correctness and efficiency objectives in online RL? -> we break the correctness-efficiency Pareto frontier! +125% relative improvement in CWM32B compared to standard RLVR (13.7 ->30.9 pass@1@top30%) +150% with Qwen32B Paper: arxiv.org/abs/2607.25970
3
3
29
4,082
Congratulations @ghita__ha ! Super super proud of you
Beyond excited to join @NotionHQ where I will be leading Model Research, alongside the whole ZeroEntropy team!
1
4
1,627
Super exciting collab! @elvisnavah @mimicrobotics and @bfl_ai , congratulations for the release
Four years ago, in the early days of my PhD, I was obsessed with Stable Diffusion, and with the work of @robrombach, @pess_r, @andi_blatt and their team in image generation. You could say I had one north star: take these generative techniques and bring them to robotics. Today, it feels unreal to announce @mimicrobotics' official collaboration with @bfl_ai and to reveal FLUX-mimic, our next-generation Video-Action Model for general purpose dexterity. The bet behind it: control reduces to visual prediction. A model that can predict how a scene will unfold has already learned the physics of manipulation. So we built our Video-Action Model architecture on top of FLUX 3, the strongest video backbone available today. We're already testing and deploying it with manufacturing leaders like @AudiOfficial, on complex manipulation long considered impossible for conventional automation. FLUX-mimic is the culmination of years of work: the promise of true multimodality extended beyond the visual, into the domain of action. A new bet on general-purpose manipulation. And we're just getting started.
11
1,637
What are the implications of this?
hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final ((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)
2
4
1,417
François Porcher retweeted
🚨Introducing Cluster with Auctions (CwA), my first work at Meta Superintelligence Labs ! What if the vectors in your database had to bid at auction for a seat in their cluster? 🔨 That's CwA — and it makes partition-based vector search up to 4× faster out-of-distribution, using as many parameters as K-Means centroids. With @hjegou, Matthijs Douze, @pierrefdz, Gergely Szilvasy
7
17
1
39
2,853
Impressive UMA, well done!
In less than 9 months, we designed and assembled a prototype of our humanoid robot, built our data pipeline, training infra, firmware, and AI stack from the ground up. The task below requires millimeter-level precision. It's a validation of the entire pipeline.
1
918
François Porcher retweeted
(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.
5,313
3,537
3,602
45,178
23,719,372
Pretty cool work which does planning in the policy space instead of action space, and with jumpy world models!
How do we get RL agents to plan over long horizons without drowning in compounding errors? Instead of reasoning over primitive, step-by-step actions, what if they could compose pre-trained policies across multiple timescales? Poster #1117 Wed 8th at 2:30 pm #ICML2026
7
1,911
9/ Results 📊 FlowWM improves World Modeling for perception tasks, mode coverage, and horizon robustness. We test this both on FuturePerception and on Bouncing Shapes, where we can enumerate all possible stochastic futures and measure precision/recall/F1.
2
1
19
2,547