Thrilled to share our new #NeurIPS2025 paper done at @GoogleDeepMind, Plasticity as the Mirror of Empowerment
We prove every agent faces a trade-off between its capacity to adapt (plasticity) and its capacity to steer (empowerment)
Paper: https://t.co/prWpkdPojb
🧵🧵🧵👇
Just published in @PNASNews, we resolve a 50-year-old riddle from Richard Feynman's handwritten notes, prove and generalize it, and run a large-scale human study to reveal near-optimal heuristics in sequential decision problems:
https://t.co/4AOM1iDqG2
"Imperfect World Models are Exploitable"
World models can look accurate, but still rank policies incorrectly, saying policy A is better than policy B when the real environment says the reverse.
This paper formalizes that failure as model exploitation and proves it is basically unavoidable for any nontrivial, nonequivalent world model on broad policy sets.
It also connects this to reward hacking and derives a safe horizon showing how model error compounds with planning depth.
Reminder that the deadline for the Finding the Frame workshop @RL_Conference is coming up on May 22 AoE ⏰
We're excited to read your submissions reflecting on the philosophy, practice, and formalisms of reinforcement learning!
🔗 More details: https://t.co/mNy0SGK1H2
Finding the Frame will be back at @RL_Conference 2026 with a fantastic speaker lineup! We welcome submissions that reflect on the philosophy, practice, and formalisms of reinforcement learning.
📅Submission deadline: May 22, 2026 (AoE)
More details: https://t.co/bYbIW3QKB7
Standard RL assumes a stable world. The real world may not.
♾ Introducing the Continual RL Workshop @RL_Conference 2026, Montreal, Canada.
🤖 Agents should never stop learning!
🖼️ Site: https://t.co/ZYpBUPbRqR
📄 Submit: https://t.co/Lf5Yj55pDe
Really excited to present our recent work at #ICLR2026 this week!
We discover highly coordinated joint behaviours and integrate them into the skill sets of MARL agents, accelerating the search for effective joint strategies in downstream tasks.🧵
Paper: https://t.co/AZYQQOlHFq
New preprint of a paper with Eunice Yiu to appear in Philosophical Transactions A, Special issue: World models, 2026. The theoretical link between empowerment in RL and Bayesian causal models with cool new data. https://t.co/3sGo14sh1f
We have the keynote speakers for RLC2026 now: Thrilled to welcome @contactrika, @ravi_iitm, @SheilaMcIlraith, @marcgbellemare, and @danijarh!
Details: https://t.co/QMMeP8JSPx
The RL community is coming together this August in Montréal, Québec, Canada. Hope you make it!
The "decoupling of information and energy" is a major point of divergence between biological and artificial computers.
Brains are efficient, modern AI isn't. And energy consumption is the biggest bottleneck in scaling AI (you can't hallucinate electrons into existence).
To address this we need an "energy-aware theory of computation." And this new preprint is an attempt to address this.
[1/11] 🧵
Newton gave force, inertia, and motion precise mathematical definitions. This unlocked centuries of progress in mechanics.
Turing did it for "computation," and Shannon for "information."
Today, "agency" still feels a bit like "computation" before Turing: everybody uses the word, but there is no widely accepted precise definition.
@dabelcs argue that RL needs to take this problem more seriously.
[1/3]🧵
PERSIST is a world model that ditches pixel-based histories for a 3D world state. Instead of searching through an ever-growing sequence of past pixel observations, PERSIST retrieves spatial information from a dynamically evolving 3D representation.
This change improves the spatial memory, 3D consistency, and long-horizon stability of the model, enabling interactive experiences within coherent and evolving 3D worlds.
#MachineLearning #WorldModels #GenerativeAI #3DComputerVision #ComputerVision #Genie3 #AI
Assembling a team at DeepMind in London.
Scaling up RL for post-training is working, but right now it's still mostly hacks and dark arts (pretraining circa 2019).
Pre-training wasn't always scaling laws and log-log plots; someone had to find the simplicity.
We aim to do the same.
If you're interested in doing things right in a research-first environment that scales all the way, please apply: https://t.co/rZZPa9PRn7
Hiring a postdoc for the Normativity Lab at Johns Hopkins (2026 start). Looking for multiagent systems expertise (RL/generative agents) + interdisciplinary background in AI and cognitive science/econ/cultural evolution.
https://t.co/RTcXrIu9gE
New preprint in advance of a Phil Trans paper. Outlining a theoretical argument bridging Bayesian causal learning and empowerment in reinforcement learning. And empirical data that kids do too!
https://t.co/3sGo14sOQN