Why does in-context learning fail sometimes? Evaluating in-context learning on open and closed questions

TL;DR

This paper evaluates in-context learning performance on open and closed questions, revealing that more relevant contexts do not always improve results, especially for difficult or novel questions, highlighting the need for better evaluation methods.

Contribution

The study introduces a new benchmark for assessing in-context learning on scientific questions and uncovers the nuanced effects of context relevance on model performance.

Findings

More relevant context does not always improve performance.

Open questions and high-difficulty questions are more affected by context relevance.

Context selection for LLMs is highly application-dependent.

Abstract

We measure the performance of in-context learning as a function of task novelty and difficulty for open and closed questions. For that purpose, we created a novel benchmark consisting of hard scientific questions, each paired with a context of various relevancy. We show that counter-intuitively, a context that is more aligned with the topic does not always help more than a less relevant context. This effect is especially visible for open questions and questions of high difficulty or novelty. This result reveals a fundamental difference between the treatment of close-form and open-form questions by large-language models and shows a need for a more robust evaluation of in-context learning on the variety of different types of questions. It also poses a new question of how to optimally select a context for large language models, especially in the context of Retrieval Augmented Generation (RAG) systems. Our results suggest that the answer to this question can be highly application-dependent and might be contingent on factors including the format of the question, the perceived difficulty level of the questions, and the novelty or popularity of the information we seek.