Evolution Strategies as a Scalable Alternative to Reinforcement Learning
OpenAI published research showing that evolution strategies (ES), a decades-old optimization technique, can match standard reinforcement learning performance on benchmarks like Atari and MuJoCo. The approach offers practical advantages over RL including easier parallelization and fewer hyperparameter sensitivities. This positions ES as a viable alternative training paradigm for policy optimization tasks.
Related guides (3)
Related events (8)
Evolved Policy Gradients: OpenAI Meta-Learning via Loss Function Evolution
OpenAI released Evolved Policy Gradients (EPG), a meta-learning method that evolves the loss function used to train reinforcement learning agents rather than hand-designing it. The approach enables faster adaptation to novel tasks, with agents demonstrating generalization to test-time scenarios outside their training distribution, such as navigating to objects placed in new locations. EPG represents an experimental direction in automated algorithm discovery for RL.
Some considerations on learning to explore via meta-reinforcement learning
OpenAI published a research post examining exploration strategies learned through meta-reinforcement learning. The work investigates how agents can acquire exploration behaviors through meta-learning rather than having them hand-designed. This is an early OpenAI contribution to the intersection of meta-learning and RL, predating the current frontier model era.
Learning Montezuma's Revenge from a Single Demonstration
OpenAI trained a reinforcement learning agent to achieve a score of 74,500 on Montezuma's Revenge using a single human demonstration, surpassing all previously published results. The method is straightforward: the agent plays episodes starting from carefully selected states drawn from the demonstration, optimizing game score via PPO. This approach demonstrates that imitation-seeded curriculum learning can dramatically improve exploration in hard-exploration environments. The same PPO algorithm underpins OpenAI Five.
Reptile: A Scalable Meta-Learning Algorithm from OpenAI
OpenAI introduced Reptile, a meta-learning algorithm that works by repeatedly sampling tasks, running stochastic gradient descent, and updating initial parameters toward the task-specific learned parameters. It is mathematically related to first-order MAML but requires only black-box access to standard optimizers like SGD or Adam. The algorithm is positioned as computationally efficient and comparably performant to MAML-based approaches.
Evolution through large models
OpenAI published a blog post titled 'Evolution through large models' in June 2022, exploring the relationship between large-scale models and evolutionary or emergent capabilities. The post appears to examine how scaling laws and large model training relate to the emergence of novel behaviors and capabilities. As a Tier 1 source publication from OpenAI, it likely addresses foundational themes around capability emergence in large language models.
Vector Policy Optimization: Training for Diversity Improves Test-Time Search
Vector Policy Optimization (VPO) is a new RL post-training algorithm for LLMs that replaces the scalar reward paradigm with vector-valued rewards, explicitly training models to produce diverse solution sets that specialize across different reward trade-offs. VPO is designed as a near-drop-in replacement for the GRPO advantage estimator and targets inference-scaling search procedures like AlphaEvolve. Across four tasks, VPO matches or outperforms scalar RL baselines on pass@k and best@k metrics, with advantages growing as search budget increases, and unlocks evolutionary search problems that GRPO-trained models cannot solve. The paper argues that diversity-optimized post-training may need to become the default as inference-time search becomes standard.
SAERL: Using Sparse Autoencoders to Guide LLM Reinforcement Learning Data Engineering
SAERL is a post-training data engineering framework that uses Sparse Autoencoders (SAEs) — a mechanistic interpretability tool — to extract intrinsic model signals for controlling data diversity, difficulty, and quality during RL fine-tuning. The framework applies SAE-space clustering for batch diversity, a difficulty proxy for curriculum ordering, and a quality probe for data filtering. On Qwen2.5-Math-1.5B with GRPO, SAERL achieves 3% average accuracy improvement and reaches target accuracy with 20% fewer training steps. SAE representations transfer across model families and scales, suggesting broad applicability as a lightweight data engineering tool.
Benchmarking Safe Exploration in Deep Reinforcement Learning
OpenAI published a benchmark for evaluating safe exploration in deep reinforcement learning, addressing the challenge of training agents that avoid unsafe behaviors during the learning process. The work provides standardized environments and metrics to measure how well RL algorithms constrain harmful actions while still achieving task objectives. This is an early contribution to the safety-aware RL research area, predating more recent alignment-focused work.