NeurIPS-2025

\( \newcommand{\states}{\mathcal{S}} \newcommand{\actions}{\mathcal{A}} \newcommand{\observations}{\mathcal{O}} \newcommand{\rewards}{\mathcal{R}} \newcommand{\traces}{\mathbf{e}} \newcommand{\transition}{P} \newcommand{\reals}{\mathbb{R}} \newcommand{\naturals}{\mathbb{N}} \newcommand{\complexs}{\mathbb{C}} \newcommand{\field}{\mathbb{F}} \newcommand{\numfield}{\mathbb{F}} \newcommand{\expected}{\mathbb{E}} \newcommand{\var}{\mathbb{V}} \newcommand{\by}{\times} \newcommand{\partialderiv}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\defineq}{\stackrel{{\tiny\mbox{def}}}{=}} \newcommand{\defeq}{\stackrel{{\tiny\mbox{def}}}{=}} \newcommand{\eye}{\Imat} \newcommand{\hadamard}{\odot} \newcommand{\trans}{\top} \newcommand{\inv}{{-1}} \newcommand{\argmax}{\operatorname{argmax}} \newcommand{\Prob}{\mathbb{P}} \newcommand{\avec}{\mathbf{a}} \newcommand{\bvec}{\mathbf{b}} \newcommand{\cvec}{\mathbf{c}} \newcommand{\dvec}{\mathbf{d}} \newcommand{\evec}{\mathbf{e}} \newcommand{\fvec}{\mathbf{f}} \newcommand{\gvec}{\mathbf{g}} \newcommand{\hvec}{\mathbf{h}} \newcommand{\ivec}{\mathbf{i}} \newcommand{\jvec}{\mathbf{j}} \newcommand{\kvec}{\mathbf{k}} \newcommand{\lvec}{\mathbf{l}} \newcommand{\mvec}{\mathbf{m}} \newcommand{\nvec}{\mathbf{n}} \newcommand{\ovec}{\mathbf{o}} \newcommand{\pvec}{\mathbf{p}} \newcommand{\qvec}{\mathbf{q}} \newcommand{\rvec}{\mathbf{r}} \newcommand{\svec}{\mathbf{s}} \newcommand{\tvec}{\mathbf{t}} \newcommand{\uvec}{\mathbf{u}} \newcommand{\vvec}{\mathbf{v}} \newcommand{\wvec}{\mathbf{w}} \newcommand{\xvec}{\mathbf{x}} \newcommand{\yvec}{\mathbf{y}} \newcommand{\zvec}{\mathbf{z}} \newcommand{\Amat}{\mathbf{A}} \newcommand{\Bmat}{\mathbf{B}} \newcommand{\Cmat}{\mathbf{C}} \newcommand{\Dmat}{\mathbf{D}} \newcommand{\Emat}{\mathbf{E}} \newcommand{\Fmat}{\mathbf{F}} \newcommand{\Gmat}{\mathbf{G}} \newcommand{\Hmat}{\mathbf{H}} \newcommand{\Imat}{\mathbf{I}} \newcommand{\Jmat}{\mathbf{J}} \newcommand{\Kmat}{\mathbf{K}} \newcommand{\Lmat}{\mathbf{L}} \newcommand{\Mmat}{\mathbf{M}} \newcommand{\Nmat}{\mathbf{N}} \newcommand{\Omat}{\mathbf{O}} \newcommand{\Pmat}{\mathbf{P}} \newcommand{\Qmat}{\mathbf{Q}} \newcommand{\Rmat}{\mathbf{R}} \newcommand{\Smat}{\mathbf{S}} \newcommand{\Tmat}{\mathbf{T}} \newcommand{\Umat}{\mathbf{U}} \newcommand{\Vmat}{\mathbf{V}} \newcommand{\Wmat}{\mathbf{W}} \newcommand{\Xmat}{\mathbf{X}} \newcommand{\Ymat}{\mathbf{Y}} \newcommand{\Zmat}{\mathbf{Z}} \newcommand{\Sigmamat}{\boldsymbol{\Sigma}} \newcommand{\identity}{\Imat} \newcommand{\epsilonvec}{\boldsymbol{\epsilon}} \newcommand{\thetavec}{\boldsymbol{\theta}} \newcommand{\phivec}{\boldsymbol{\phi}} \newcommand{\muvec}{\boldsymbol{\mu}} \newcommand{\sigmavec}{\boldsymbol{\sigma}} \newcommand{\jacobian}{\mathbf{J}} \newcommand{\ind}{\perp!!!!\perp} \newcommand{\bigoh}{\text{O}} \)

BlueSky/LinkedIn posts:

It has been a blast re-engaging with new and old friends in RL/ML/AI at #NeurIPS2025! It has been amazing to see how much the community has grown significantly since I participated first (at NeurIPS2016, Barcelona!). The excitement towards building a better world through AI has never been more present, and the reality of the consequences of our actions has hit the community deeply. The common theme this year has been a call for multi-disciplinary collaboration to deeply engage with applied domains, improve life for all not just some, to better understand how our most impressive models reason so we can make progress, and (in RL especially) rethinking and recalibrating to be more productive for the future.

There have been so many great discussions and posters that it is hard to choose a few, but here are some of my highlights this year which follow this theme:

Multi-disciplinary collaboration was called for by almost all of the Keynote speakers.

  • Yejin Choi’s excellent framing of the SFT vs. RL conversation in LLMs
  • Melanie Mitchell’s reminder that we must be creative in uncovering why our methods behave like they do
  • Aligning Theorists and Practitioners in RL workshop
  • Sejnowski-Hinton Award: Hinton mentioned his most fulfilling experience as a researcher was his multi-disciplinary collaboration with Sejnowski, and the new reward is incentivizing the creation of new collaborations working to uncover the mysteries of the brain.

Improve life for all not just some:

  • Zeynep Tufekci: where should we be looking for the actual disruption our technology will have?
  • Yashua Bengio’s call for major safety ideas: A Safety Argument for the Scientist AI

to better understand how our most impressive models reason so we can make progress

rethinking and recalibrating to be more productive for the future

Code:

  • Unifloral
  • Fast RL Tools
  • Gymnasium

Ideas:

  • Jossiah Hannah’s Paper
  • Counterintuitive (minimize Q-learning)

One idea that kept popping up was action chunking! A really nice paper was the MBRL imporovements w/ action chunking, where they can do planning in continuous time! It is interesting how temporal abstractions have popped up almost indepedently from options again. It is clear abstraction in the action space as well as the state/observation space is improtent for the agent to attend to the most important degrees of freedom.