Juliette Decugis @ ICML🇰🇷

What advantage to use, and when? Everyone's proposing new advantage functions for RL with LLMs but nobody knows why they work or fail. We break this down and build FADE a self-adapting advantage to get +14% on LiveCodeBench v6 in 40% less steps. Paper: arxiv.org/pdf/2607.01490

🇰🇷 #ICML2026 Alert! 🇰🇷 Come check out our new work with @mathuvu_ and @ylecun 👀 📍 Starting 2PM - Poster #1606 💡 Representation learning for text done efficiently - clean scaling, 2× retrieval, ~100× less compute. 🚨 Spoiler Alert: BERT does not scale

arxiv.org/pdf/2607.02390 Really cute work from @AIatMeta’s CodeGen team in DecompRL, most notably @DecugisJuliette and @FabianGloeckle You train a code model first to decompose a problem in multiple functions, then implement each of those -> through permutations of the rollouts of the implementations, you can actually size a diverse code training data much easier They make a specific policy gradient algorithm to grade that process, first training the decomposition then the implementation policy using a logmeanexp function to compute the objective of the multiple rollouts and aggregating over a leave-one-out baseline Models trained are as good as other RL algorithms… but work better on larger token budget, and most notably have much less emphasis on GPU (less need for rollouts) and much more emphasis on CPU (more need for function verification) A pretty cool work that makes you think about domain specific training! 😊

I’m at ICML🇰🇷 this week in Seoul ! Feel free to reach out if you want to chat about RL and codegen! Will also present @DL4Code my work on advantages and at the RLxF workshop work on efficient codegen (led by @PierreChambon6)!

@PengmingWang Yes agreed but here difficulty is defined relative to the current policy. With a fixed dataset, early in training you want to maximize learning signal on solvable hard tasks. As the policy improves, there are less hard tasks while medium tasks provide denser, reliable updates.

@DecugisJuliette Feels counter intuitive, no? Wouldn't I get more & less variance signal if I attempt hard tasks later, as solve rates rise?

What advantage to use, and when? Everyone's proposing new advantage functions for RL with LLMs but nobody knows why they work or fail. We break this down and build FADE a self-adapting advantage to get +14% on LiveCodeBench v6 in 40% less steps. Paper: arxiv.org/pdf/2607.01490

Joint work with @seano_research and my amazing advisors: @BachFrancis , @syhw , @TacoCohen We hope this gives the community a way to sort through the advantage literature and opens the door to dynamic allocation of gradient mass. 🚀

Open questions we're excited about: - how can we extend this framework to noisy verifiers? (eg, with judge models) - can we get partial credit from failures to make failure-only RL learn?

🧠⌨️ Decode language from brain activity without surgery. 🧠⌨️ Brain2Qwerty V1 is officially published in Nature Neuroscience. Today, we're releasing Brain2Qwerty V2. We achieve unprecedented performance for a non-invasive MEG setup. Details below 🧵👇

Amazing work led by @KunhaoZ Turns out your model either optimises speed or code correctness during RL and you can minimise both failures by extrapolating between RL checkpoints! Let's see how we can bring this to training 🧐

@arnal_charles @KempeLab Congrats well deserved !

Very excited to be joining Meta Superintelligence Labs as a Research Scientist! I’ll be continuing my work on RL and AI for maths with @KempeLab, Rémi Munos, and my longtime partner in crime, Vivien Cabannes.

(1/9) Experience replay can cut LLM RL training compute by up to ~40% (without hurting final accuracy—and sometimes improving it). Paper: arxiv.org/abs/2604.08706

Can LLMs reliably predict program termination? We evaluate frontier LLMs in the International Competition on Software Verification (SV-COMP) 2025, directly competing with state-of-the-art verification systems. @AIatMeta @HebrewU @Bloomberg @imperialcollege @ucl @jordiae @pascalkesseli @jvanegue @HyadataLab @adiyossLC @PeterOHearn12 Paper: arxiv.org/pdf/2601.18987 Website: orensultan.com/llms_halting_p… 🧵👇 1/n

TL;DR: I made a Transformer that conditions its generation on latent variables. To do so an encoder Transformer only needs a source of randomness during generation, but then it needs an encoder for training, as a [conditional] VAE. 1/5

Checkout 32B CWM from our team❤️ Releasing 3 model checkpoints to empower research: pre-trained -> SFT -> post trained 🚀 Super proud to have helped the post training efforts learning tons from all these amazing researchers!