Towards Reasoning in Large Language Models: A Survey - Summary

Arxiv URL: https://arxiv.org/abs/2212.10403v2

Authors: Jie Huang, Kevin Chen-Chuan Chang

Summary:

This paper provides a comprehensive overview of the current state of knowledge on reasoning in Large Language Models (LLMs), including techniques for improving and eliciting reasoning in these models, methods and benchmarks for evaluating reasoning abilities, findings and implications of previous research in this field, and suggestions on future directions.

Key Insights & Learnings:

• Reasoning is a fundamental aspect of human intelligence that plays a crucial role in activities such as problem solving, decision making, and critical thinking.
• Large language models (LLMs) have made significant progress in natural language processing, and there is observation that these models may exhibit reasoning abilities when they are sufficiently large.
• Despite the strong performance of LLMs on certain reasoning tasks, it remains unclear whether LLMs are actually reasoning and to what extent they are capable of reasoning.
• Techniques for enhancing/eliciting reasoning in LLMs include Fully Supervised Finetuning, Prompting & In-Context Learning, Chain of Thought and Its Variants, Rationale Engineering, Problem Decomposition, and others.
• Future research directions include exploring hybrid methods, reasoning-enhanced training & prompting, bootstrapping & self-improving, and developing better evaluation & analysis methods.

Terms Mentioned: Reasoning, Large Language Models, Natural Language Processing, Deductive Reasoning, Inductive Reasoning, Abductive Reasoning, Formal Reasoning, Informal Reasoning, Fully Supervised Finetuning, Prompting & In-Context Learning, Chain of Thought, Rationale Engineering, Problem Decomposition, Hybrid Method, Reasoning-Enhanced Training & Prompting, Bootstrapping & Self-Improving, Evaluation & Analysis

Technologies / Libraries Mentioned: GPT-3, BERT