Language-based Trial and Error Falls Behind in the Era of Experience

TL;DR

This paper introduces SCOUT, a framework that uses lightweight probes to efficiently explore environments and fine-tune large language models, significantly improving their performance on unseen, nonlinguistic tasks while reducing computational costs.

Contribution

The paper presents SCOUT, a novel exploration framework that decouples exploration from exploitation, enabling large language models to better handle unseen tasks with less computational expense.

Findings

SCOUT improves LLM performance on unseen tasks.

SCOUT reduces GPU hours by about 60%.

Qwen2.5-3B-Instruct achieves 0.86 score with SCOUT.

Abstract

While Large Language Models (LLMs) excel in language-based agentic tasks, their applicability to unseen, nonlinguistic environments (e.g., symbolic or spatial tasks) remains limited. Previous work attributes this performance gap to the mismatch between the pretraining distribution and the testing distribution. In this work, we demonstrate the primary bottleneck is the prohibitive cost of exploration: mastering these tasks requires extensive trial-and-error, which is computationally unsustainable for parameter-heavy LLMs operating in a high dimensional semantic space. To address this, we propose SCOUT (Sub-Scale Collaboration On Unseen Tasks), a novel framework that decouples exploration from exploitation. We employ lightweight "scouts" (e.g., small MLPs) to probe environmental dynamics at a speed and scale far exceeding LLMs. The collected trajectories are utilized to bootstrap the LLM via Supervised Fine-Tuning (SFT), followed by multi-turn Reinforcement Learning (RL) to activate its latent world knowledge. Empirically, SCOUT enables a Qwen2.5-3B-Instruct model to achieve an average score of 0.86, significantly outperforming proprietary models, including Gemini-2.5-Pro (0.60), while saving about 60% GPU hours consumption.