LLMs Can Reason Faster Only If We Let Them

摘要

Large language models (LLMs) are making inroads into classical AI problems such as automated planning, yet key shortcomings continue to hamper their integration. Chain-of-Thought (CoT) struggles in complex multi-step reasoning, and Tree-of-Thoughts requires multiple queries that increase computational overhead. Recently, Algorithm-of-Thoughts (AoT) have shown promise using in-context examples, at the cost of significantly longer solutions compared to CoT. Aimed at bridging the solution length gap between CoT and AoT, this paper introduces AoT-O3, which combines supervised finetuning on AoT-style plans with a reinforcement learning (RL) framework designed to reduce solution length. The RL component uses a reward model that favors concise, valid solutions while maintaining planning accuracy. Empirical evaluations indicate that AoT-O3 shortens solution length by up to 80\% compared to baseline AoT while maintaining or surpassing prior performance. These findings suggest a promising pathway for more efficient, scalable LLM-based planning.