Self-correction is a highly desirable capability of large language models (LLMs), yet it has consistently been found to be largely ineffective in modern LLMs. Current methods for training self-correction typically depend on either multiple models, a more advanced model, or additional forms of supervision. To address these shortcomings, we develop a multi-turn online reinforcement learning (RL) approach, SCoRe, that significantly improves an LLM's self-correction ability using entirely self-generated data. To build SCoRe, we first show that variants of supervised fine-tuning (SFT) on offline model-generated correction traces are insufficient for instilling self-correction behavior. In particular, we observe that training via SFT either suffers from a distribution mismatch between the training data and the model's own responses or implicitly prefers only a certain mode of correction behavior that is often not effective at test time. SCoRe addresses these challenges by training under the model's own distribution of self-generated correction traces and using appropriate regularization to steer the learning process into learning a self-correction strategy that is effective at test time as opposed to simply fitting high-reward responses for a given prompt. This regularization prescribes running a first phase of RL on a base model to generate a policy initialization that is less susceptible to collapse and then using a reward bonus to amplify self-correction during training. When applied to Gemini 1.0 Pro and 1.5 Flash models, we find that SCoRe achieves state-of-the-art self-correction performance, improving the base models' self-correction by 15.6% and 9.1% respectively on the MATH and HumanEval benchmarks.

Foundation models have shown impressive adaptation and scalability in supervised and self-supervised learning problems, but so far these successes have not fully translated to reinforcement learning (RL). In this work, we demonstrate that training an RL agent at scale leads to a general in-context learning algorithm that can adapt to open-ended novel embodied 3D problems as quickly as humans. In a vast space of held-out environment dynamics, our adaptive agent (AdA) displays on-the-fly hypothesis-driven exploration, efficient exploitation of acquired knowledge, and can successfully be prompted with first-person demonstrations. Adaptation emerges from three ingredients: (1) meta-reinforcement learning across a vast, smooth and diverse task distribution, (2) a policy parameterised as a large-scale attention-based memory architecture, and (3) an effective automated curriculum that prioritises tasks at the frontier of an agent’s capabilities. We demonstrate characteristic scaling laws with respect to network size, memory length, and richness of the training task distribution. We believe our results lay the foundation for increasingly general and adaptive RL agents that perform well across ever-larger open-ended domains.

In the absence of external rewards, agents can still learn useful behaviors by identifying and mastering a set of diverse skills within their environment. Existing skill learning methods use mutual information objectives to incentivize each skill to be diverse and distinguishable from the rest. However, if care is not taken to constrain the ways in which the skills are diverse, trivially diverse skill sets can arise. To ensure useful skill diversity, we propose a novel skill learning objective, Relative Variational Intrinsic Control (RVIC), which incentivizes learning skills that are distinguishable in how they change the agent's relationship to its environment. The resulting set of skills tiles the space of affordances available to the agent. We qualitatively analyze skill behaviors on multiple environments and show how RVIC skills are more useful than skills discovered by existing methods in hierarchical reinforcement learning.

Learned models of the environment provide reinforcement learning (RL) agents with flexible ways of making predictions about the environment.Models enable planning, i.e. using more computation to improve value functions or policies, without requiring additional environment interactions.In this work, we investigate a way of augmenting model-based RL, by additionally encouraging a learned model and value function to be jointly \emph{self-consistent}.This lies in contrast to classic planning methods like Dyna, which only update the value function to be consistent with the model.We propose a number of possible self-consistency updates, study them empirically in both the tabular and function approximation settings, and find that with appropriate choices self-consistency can be useful both for policy evaluation and control.

An agent might be said, informally, to have mastery of its environment when it has maximised the effective number of states it can reliably reach. In practice, this often means maximizing the number of latent codes that can be discriminated from future states under some short time horizon (e.g. \cite{eysenbach2018diversity}). By situating these latent codes in a globally consistent coordinate system, we show that agents can reliably reach more states in the long term while still optimizing a local objective. A simple instantiation of this idea, \textbf{E}ntropic \textbf{D}esired \textbf{D}ynamics for \textbf{I}ntrinsic \textbf{C}on\textbf{T}rol (EDDICT), assumes fixed additive latent dynamics, which results in tractable learning and an interpretable latent space. Compared to prior methods, EDDICT's globally consistent codes allow it to be far more exploratory, as demonstrated by improved state coverage and increased unsupervised performance on hard exploration games such as Montezuma's Revenge.