Thinking LLMs: General Instruction Following with Thought Generation

TL;DR

This paper introduces a training method that enables large language models to generate explicit thoughts before answering, improving their performance across reasoning and non-reasoning tasks without additional human data.

Contribution

It proposes an iterative search and optimization approach to teach LLMs explicit thinking abilities for instruction following without supervised data.

Findings

Improved performance on AlpacaEval and Arena-Hard benchmarks.

Gains observed in non-reasoning categories like marketing, health, and general knowledge.

Effective enhancement of LLMs' thinking capabilities without extra human annotations.

Abstract

LLMs are typically trained to answer user questions or follow instructions similarly to how human experts respond. However, in the standard alignment framework they lack the basic ability of explicit thinking before answering. Thinking is important for complex questions that require reasoning and planning -- but can be applied to any task. We propose a training method for equipping existing LLMs with such thinking abilities for general instruction following without use of additional human data. We achieve this by an iterative search and optimization procedure that explores the space of possible thought generations, allowing the model to learn how to think without direct supervision. For each instruction, the thought candidates are scored using a judge model to evaluate their responses only, and then optimized via preference optimization. We show that this procedure leads to superior performance on AlpacaEval and Arena-Hard, and shows gains from thinking on non-reasoning categories such as marketing, health and general knowledge, in addition to more traditional reasoning & problem-solving tasks.