Jakob Nicolaus Foerster

StochasTok: Improving Fine-Grained Subword Understanding in LLMs

High Accuracy, Less Talk (HALT): Reliable LLMs through Capability-Aligned Finetuning

The Decrypto Benchmark for Multi-Agent Reasoning and Theory of Mind

Programming by Backprop: An Instruction is Worth 100 Examples When Finetuning LLMs

DéjàQ: Open-Ended Evolution of Diverse, Learnable and Verifiable Problems

HyperVLA: Efficient Inference in Vision-Language-Action Models via Hypernetworks

SOReL and TOReL: Two Methods for Fully Offline Reinforcement Learning

Learning When to Plan: Efficiently Allocating Test-Time Compute for LLM Agents

A Clean Slate for Offline Reinforcement Learning

Kinetix: Investigating the Training of General Agents through Open-Ended Physics-Based Control Tasks

AgentBreeder: Mitigating the AI Safety Risks of Multi-Agent Scaffolds via Self-Improvement

Improving Regret Approximation for Unsupervised Dynamic Environment Generation

Simplifying Deep Temporal Difference Learning

AI Research Agents for Machine Learning: Search, Exploration, and Generalization in MLE-bench

Meta-Learning Objectives for Preference Optimization

LILO: Learning to Reason at the Frontier of Learnability

Imagined Autocurricula

Expected Return Symmetries

LOB-Bench: Benchmarking Generative AI for Finance - an Application to Limit Order Book Data

BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games

The Decrypto Benchmark for Multi-Agent Reasoning and Theory of Mind

MLGym: A New Framework and Benchmark for Advancing AI Research Agents

TICKing All the Boxes: Generated Checklists Improve LLM Evaluation and Generation

Towards Learning to Reason at Pre-Training Scale

Ad-Hoc Human-AI Coordination Challenge

ADIOS: Antibody Development via Opponent Shaping

OvercookedV2: Rethinking Overcooked for Zero-Shot Coordination

The Complexity Dynamics of Grokking

Learning Loss Landscapes in Preference Optimization

CURATe: Benchmarking Personalised Alignment of Conversational AI Assistants

Ad-Hoc Human-AI Coordination Challenge

Beyond the Boundaries of Proximal Policy Optimization

Source2Synth: Synthetic Data Generation and Curation Grounded in Real Data Sources

Investigating Online RL in World Models

LOB-Bench: Benchmarking Generative AI for Finance - with an Application to Limit Order Book Markets

Do Symbolic or Black-Box Representations Generalise Better In Learned Optimisation?