Aneesh Muppidi @aneeshers

@sean_n_wu @COLM_conf Congrats Sean!!

Excited to share that our paper introducing BAS (Behavioral Alignment Score), a decision-theoretic metric for evaluating LLM confidence, was accepted to COLM 2026! 🎉 arxiv: arxiv.org/pdf/2604.03216 @COLM_conf

keyboard_arrow_left Previous keyboard_arrow_right Next

Thank you, @OpenAI, for hosting us in San Francisco last month and for sharing our stories. During our visit, I led a breakout group with @johnohallman on AI for scientific discovery. More in the video below!

We hope you’ve enjoyed a sneak peek of work from BOLD and our collaborators at #ICML2026! See below for a full summary of where you can find us this week: ▶️ (Poster) Procedural Generation of Algorithm Discovery Tasks in Machine Learning, Hall A #1803, Tuesday 10:30 - 12:15, led by @AlexDGoldie ▶️ (Poster) h1: Bootstrapping LLMs to Reason over Longer Horizons via Reinforcement Learning, Hall A #2704, led by Alesia Ivanova @sumeetrm ▶️ (Poster) Goal-Conditioned Agents that Learn Everything All at Once, Hall A #310, Tuesday 14:00 - 15:45, led by @mitrma ▶️ (Poster) Rubric Curriculum RL: Exploiting the Generation-Verification Gap in Creative Writing, Hall A #2605, Wednesday 14:30 - 14:15, led by Tejas Krishnan @sumeetrm ▶️ (Poster) Evolution Strategies at the Hyperscale, Hall A #3712, Thursday 10:30 - 12:15, led by @bidiptas13 @JuanDuquevan Mattie Fellows ▶️ (Poster) Dreaming in Code for Curriculum Learning in Open-Ended Worlds, Hall A #213, Wednesday 17:00 - 18:45, led by @k_mitsides ▶️ (Poster) Evolution Strategies at the Hyperscale, Hall A #3712, Thursday 10:30 - 12:15, led by @bidiptas13 @JuanDuquevan Mattie Fellows ▶️ (Poster) The Decrypto Benchmark for Multi-Agent Reasoning and Theory of Mind, Hall A #3504, Thursday 14:30 - 16:15, led by @_andreilupu ▶️ (Poster) LongCoT: Benchmarking Long-Horizon Chain-of-Thought Reasoning, Hall A #1705, Thursday 14:30 - 16:15, led by @sumeetrm @DanielNichols10 @CharlieLondon02 Peggy Li Fabio Pizzati ▶️ (Talk) Superhuman Scientific Discovery, RLxF Worskhop, Friday 15:30 - 16:00, by @robertarail ▶️ (Panel) RLxF Worskhop, Friday 16:00 - 17:00, by @robertarail ▶️ (Workshop) Amortising Bayesian Experimental Design for Sequential Information Gathering in LLMs, FoGen Workshop, Friday, led by @jakobhartmann99 James Harvey Jhonathan Navott ▶️ (Workshop) Elicitation Format Drives Divergent LLM Geopolitical Forecasts, AI Forecasting Workshop, Saturday, led by @hariharansuhas @michalbravansky ▶️ (Workshop) EGGROLL-IPO: Pluralistic Alignment via Decentralised Post-Training with Population Preferences, Pluralistic Alignment Workshop, Saturday, led by @alfie_lamerton ▶️ (Workshop Spotlight) Abstraction for Offline Goal-Conditioned Reinforcement Learning, DEMO Workshop, Saturday, led by @ClarisseWibault

@JagdeepBhatia8 @ajwagenmaker @verityw_ @svlevine Awesome work @JagdeepBhatia8!

@simonramstedt Thanks - your work was really informative and gave us a lot to build on!

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

Introducing HABIT — a large-scale robot manipulation dataset for human-present environments, where a person shares the workspace and interacts with the robot in every episode. 60 tasks · 10,563 episodes · 164 hours of rich human-robot interaction. Toward robots that are not just capable, but safe and socially compatible around people. habit-dataset.github.io 🧵[1/7]

Now you can train robots on any preference you can describe in words (speed, precision, "delicateness") and even steer them at test time without retraining! Check out this awesome project from @marceltornev, @anubhamahajan01 and @AbhijnyaBhat!

Marcel Torné @marceltornev

2 weeks ago

We should stop optimizing robot policies against a single overall reward. Trajectories differ along many axes, such as speed, precision, and subtask completion, and one can be better on some while worse on others. If we collapse all of that into a single overall axis we lose this

10 31 155 33K 76

We should stop optimizing robot policies against a single overall reward. Trajectories differ along many axes, such as speed, precision, and subtask completion, and one can be better on some while worse on others. If we collapse all of that into a single overall axis we lose this structure making the reward ambiguous and harder to optimize. Blog: freeform-pl.github.io/fpl.website/ Paper: arxiv.org/abs/2606.32027

introducing hierarchical implicit q-chunking (HIQC) long-horizon offline reinforcement learning is hard the more steps an agent takes, the more value estimation errors compound i reduce this by decomposing the horizon into smaller segments with subgoals and action chunks

@Anqinator @FirasBDarwish Thanks Anqi!!

What I love about Aneesh’s work is how cohesive his research narrative is! I remember him presenting TRAC at NeurIPS a few years ago and this work similarly has such elegant framing and interventions. Great work!

Aneesh Muppidi @aneeshers

2 weeks ago

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

17 83 716 75K 634

@ClarisseWibault Thanks @ClarisseWibault !!

Some great work addressing an important question- I really enjoyed this paper!

Aneesh Muppidi @aneeshers

2 weeks ago

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

17 83 716 75K 634

Scaling this is what gets us to GPT 6

Aneesh Muppidi @aneeshers

2 weeks ago

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

17 83 716 75K 634

this is the type of thinking we need. our ai is too rigid.

Aneesh Muppidi @aneeshers

2 weeks ago

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

17 83 716 75K 634

@JustinWaugh Thanks Justin! We have some results on comparing across different GPUs, but agree would love to benchmark all of these ideas!

RL environments typically wait for an agent to make their decision. We challenge this assumption and introduce real-time RL: a setting where the environment continues to evolve during decision-making, and agents learn when to think longer versus act sooner. 🧵

Aneesh Muppidi @aneeshers

2 weeks ago

Introducing Real-time RL. In the real world, time isn't free. The environment keeps "moving" even when you're computing your next action. We show how RL agents can learn to adaptively think in real-time games. 1/🧵

17 83 716 75K 634

Can diffusion learn physical dynamics directly in 3D space? Yes! Introducing 🫧PhysiFormer🫧 — a unified model that generates physically plausible multi-object, multi-material 4D dynamics in world space, conditioned on initial per-vertex positions and velocities. 🧵1/8