CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay

TL;DR

CodeIt introduces a novel self-improvement method for language models tackling reasoning tasks, using prioritized hindsight replay and program relabeling to enhance performance on the ARC benchmark, achieving state-of-the-art results.

Contribution

The paper presents a scalable neuro-symbolic approach that effectively addresses reward sparsity in program synthesis, enabling successful generalization on the full ARC dataset.

Findings

Solves 15% of ARC tasks, setting a new state-of-the-art

Outperforms existing neural and symbolic baselines

Demonstrates effective inter-task generalization

Abstract

Large language models are increasingly solving tasks that are commonly believed to require human-level reasoning ability. However, these models still perform very poorly on benchmarks of general intelligence such as the Abstraction and Reasoning Corpus (ARC). In this paper, we approach ARC as a programming-by-examples problem, and introduce a novel and scalable method for language model self-improvement called Code Iteration (CodeIt). Our method iterates between 1) program sampling and hindsight relabeling, and 2) learning from prioritized experience replay. By relabeling the goal of an episode (i.e., the target program output given input) to the realized output produced by the sampled program, our method effectively deals with the extreme sparsity of rewards in program synthesis. Applying CodeIt to the ARC dataset, we demonstrate that prioritized hindsight replay, along with pre-training and data-augmentation, leads to successful inter-task generalization. CodeIt is the first neuro-symbolic approach that scales to the full ARC evaluation dataset. Our method solves 15% of ARC evaluation tasks, achieving state-of-the-art performance and outperforming existing neural and symbolic baselines. Our code is available at https://github.com/Qualcomm-AI-research/codeit .