Iterative Deployment Improves Planning Skills in LLMs

TL;DR

Iterative deployment of LLMs, involving user-curated data, enhances planning skills and induces emergent generalization, functioning similarly to reinforcement learning without explicit reward signals.

Contribution

This paper introduces a novel iterative deployment mechanism that improves LLM planning abilities and provides a theoretical link to reinforcement learning, highlighting safety and training implications.

Findings

Models show significant planning skill improvements after iterative deployment.

Later models discover longer, more complex plans than initial models.

The mechanism acts as an implicit reinforcement learning process.

Abstract

We show that iterative deployment of large language models (LLMs), each fine-tuned on data carefully curated by users from the previous models' deployment, can significantly change the properties of the resultant models. By testing this mechanism on various planning domains, we observe substantial improvements in planning skills, with later models displaying emergent generalization by discovering much longer plans than the initial models. We then provide theoretical analysis showing that iterative deployment effectively implements reinforcement learning (RL) training in the outer-loop (i.e. not as part of intentional model training), with an implicit reward function. The connection to RL has two important implications: first, for the field of AI safety, as the reward function entailed by repeated deployment is not defined explicitly, and could have unexpected implications to the properties of future model deployments. Second, the mechanism highlighted here can be viewed as an alternative training regime to explicit RL, relying on data curation rather than explicit rewards.