On the Role of Model Prior in Real-World Inductive Reasoning

TL;DR

This paper investigates how model priors influence hypothesis generation in large language models during real-world inductive reasoning, revealing that priors dominate over demonstrations and are difficult to override.

Contribution

It systematically evaluates the impact of model priors versus demonstrations across multiple tasks and models, highlighting the dominant role of priors in hypothesis generation.

Findings

Model priors primarily drive hypothesis generation.

Removing demonstrations minimally affects hypothesis quality.

Priors are resistant to override even with label flipping.

Abstract

Large Language Models (LLMs) show impressive inductive reasoning capabilities, enabling them to generate hypotheses that could generalize effectively to new instances when guided by in-context demonstrations. However, in real-world applications, LLMs' hypothesis generation is not solely determined by these demonstrations but is significantly shaped by task-specific model priors. Despite their critical influence, the distinct contributions of model priors versus demonstrations to hypothesis generation have been underexplored. This study bridges this gap by systematically evaluating three inductive reasoning strategies across five real-world tasks with three LLMs. Our empirical findings reveal that, hypothesis generation is primarily driven by the model's inherent priors; removing demonstrations results in minimal loss of hypothesis quality and downstream usage. Further analysis shows the result is consistent across various label formats with different label configurations, and prior is hard to override, even under flipped labeling. These insights advance our understanding of the dynamics of hypothesis generation in LLMs and highlight the potential for better utilizing model priors in real-world inductive reasoning tasks.