Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models
Formalizes self-improvement around the generation-verification gap, the gain a model gets from filtering its own output with its own judgment. It gives the working rule for this collection, which is that a system improves itself only where it verifies better than it generates.
Ask this paper
The gap widens predictably with pretraining compute.
Read it before designing any self-improvement loop.
Abstract
Self-improvement is a mechanism in Large Language Model (LLM) pre-training, post-training and test-time inference. We explore a framework where the model verifies its own outputs, filters or reweights data based on this verification, and distills the filtered data. Despite several empirical successes, a fundamental understanding is still lacking. In this work, we initiate a comprehensive, modular and controlled study on LLM self-improvement. We provide a mathematical formulation for self-improvement, which is largely governed by a quantity which we formalize as the generation-verification gap. Through experiments with various model families and tasks, we discover a scaling phenomenon of self-improvement -- a variant of the generation-verification gap scales monotonically with the model pre-training flops. We also examine when self-improvement is possible, an iterative self-improvement procedure, and ways to improve its performance. Our findings not only advance understanding of LLM self-improvement with practical implications, but also open numerous avenues for future research into its capabilities and boundaries.