Antoine Moulin

@jsuarez If you look at go-explore you may also want to check out arxiv.org/abs/2603.22273 too!

I am attempting to solve curriculum learning in RL in the next 2 weeks. Join me every day for 6-8 hours of livestreamed research, a 10 km run, heavy compounds, and a 1,000 calorie cut. Details + ground rules on day 1!

@_arohan_ @adinapak_ Breakfast at dishoom is underrated

@adinapak_ Both are restaurants in london. Highly recommend them

How do I buy spv on dishoom and hoppers?

But if AI mathematics continues to progress at anything like its current rate -- which is what I expect to happen -- then we will face a crisis very soon, and mathematics departments, who owe a duty of care to their students, should be urgently preparing for it.

The future of Math is mathematicians and AI agents working together. Very pleased to introduce @GoogleDeepMind's AI co-mathematician: a multi-agent system designed to actively collaborate with human experts on open-ended research mathematics. Mathematicians testing the agent across areas as diverse as group theory, Hamiltonian systems, and algebraic combinatorics have reported impressive results. In autonomous mode evaluation on the rigorous FrontierMath Tier 4 problems, AI co-mathematician scored an unprecedented 48% — a new high score among all AI systems evaluated.

@hbouammar @icmlconf When I search for the inverse RL one I only find another paper with the same title :(

ICML +3 🤙🏻🤙🏻 @icmlconf

🚨Very excited to see our work on warmth & sycophancy in LLMs out in @Nature today!🚨 We study what happens when LLMs are fine-tuned to be warmer, and find that warmth and sycophancy can be linked, with warm models showing higher errors on a range of benchmarks (🔗s below)

@lujainmibrahim @Nature @oiioxford congrats!!

Excited to share that our lab will present two Orals at the ICLR SPOT workshop this Monday: • Maximum Likelihood Reinforcement Learning (10:10–10:20) — 🏆 Best Paper Award • Expanding the Capabilities of Reinforcement Learning via Text Feedback (10:20–10:30) — Oral + 🏆 Outstanding Paper Award at LLA Workshop Come and say hi!

@Zanette_ai Congrats to the authors!!

I'm at #ICLR2026 presenting a poster 04/23! We all want to control GenAI models, but we lack tools to properly evaluate the limits of control. Here, we introduce algorithms to rigorously estimate controllable sets of any GenAI model with guarantees. Work from interning @Apple

Excited about a couple of papers of ours in ICLR this year (both in Poster Session 1 Pavilion 3 & Oral Session 2B tomorrow): (1) Sequences of Logits Reveal the Low-Rank Structure of Language Models (joint w/ @axliu42 & @AShettyV) arxiv.org/pdf/2510.24966. 1/n

@nanjiang_cs looking forward to it!!

Below is our “textbook” understanding of OPE with value-function approximation. Turns out some of them are not quite right/superficial; guess which ones need to go? ALL OF THEM!! IMSI Talk on Wed (not at ICLR): 1 ICLR paper + 1 preprint imsi.institute/activities/the…

I will be at ICLR 🇧🇷 this week and give some oral presentations and a panel discussion during the workshop days. Happy to discuss anything about RL (LLMs, robotics, theory).

@g_k_swamy congrats!!

it took a minute, but i'm proud to share that i'm finally "Dr. Swamy" :)

@Fabien_Mikol @gabrielpeyre ah j'avais pas vu !

@antoine_mln @gabrielpeyre Il y avait eu aussi Oudeyer ! x.com/Fabien_Mikol/s…

Virginie Bonnaillie-Noël (ENS) et @gabrielpeyre : on a "très clairement une rupture", des post-docs rapportant que leur thèse peut désormais être faite par l'IA en quelques minutes plutôt qu'en 3 ans ; "les performances sont assez incroyables", et "ça change incroyablement vite".

legend

Through what mechanisms can reasoning models learn faster by choosing what problems to train on, and what are the limits? Part I of a new series: "Learning to Reason with Curriculum", where we explore algorithmic principles for overcoming the limitations of pre-trained models and data. w/ Audrey Huang (@auddery), Miro Dudik (@MiroDudik), Rob Schapire, Dylan Foster (@canondetortugas) and Akshay Krishnamurthy. [1/12]

@andreamichi @Mononofu Congrats Andrea!!

depthfirst has raised an $80M Series B at a $580M valuation. Attackers are using AI to break into systems faster than ever before. depthfirst is on a mission to stop this. RT + Comment “depthfirst” and I’ll send you a FREE vibe coding security agent.

@Yadkori This also means @EurIPSConf is subject to the same rule now that it’s an official satellite event… Surprised the organizers agreed to this!

Very disappointing. That’s one less area chair responsibility for me. If I hadn’t already committed to colleagues, I wouldn’t submit a paper this year either.