New paper: Thought-Like-Pro: Enhancing Reasoning of Large Language Models through

[maykcaldas]

Paper: Thought-Like-Pro: Enhancing Reasoning of Large Language Models through

Authors: Xiaoyu Tan (1), Yongxin Deng (2), Xihe Qiu (2), Weidi Xu (1), Chao Qu

Abstract: Large language models (LLMs) have shown exceptional performance asgeneral-purpose assistants, excelling across a variety of reasoning tasks. Thisachievement represents a significant step toward achieving artificial generalintelligence (AGI). Despite these advancements, the effectiveness of LLMs oftenhinges on the specific prompting strategies employed, and there remains a lackof a robust framework to facilitate learning and generalization across diversereasoning tasks. To address these challenges, we introduce a novel learningframework, THOUGHT-LIKE-PRO In this framework, we utilize imitation learning toimitate the Chain-of-Thought (CoT) process which is verified and translatedfrom reasoning trajectories generated by a symbolic Prolog logic engine. Thisframework proceeds in a self-driven manner, that enables LLMs to formulaterules and statements from given instructions and leverage the symbolic Prologengine to derive results. Subsequently, LLMs convert Prolog-derived successivereasoning trajectories into natural language CoT for imitation learning. Ourempirical findings indicate that our proposed approach substantially enhancesthe reasoning abilities of LLMs and demonstrates robust generalization acrossout-of-distribution reasoning tasks.

Link: https://arxiv.org/abs/2407.14562

Reasoning: Reasoning: Let's think step by step in order to determine if the paper is about a language model. We start by examining the title and abstract. The title mentions "Enhancing Reasoning of Large Language Models," which directly indicates that the paper involves language models. The abstract further elaborates on the performance of large language models (LLMs) in reasoning tasks and introduces a novel framework to improve their reasoning abilities. The focus on LLMs and their reasoning capabilities strongly suggests that the paper is about language models.