Counting Hypothesis: Potential Mechanism of In-Context Learning

2026-02-02

AI Summary

This work addresses the limited understanding of in-context learning (ICL) mechanisms in large language models, which hinders error diagnosis and controllability. We propose the “counting hypothesis,” positing that models implement ICL by encoding the frequency of examples within the context, and present the first systematic validation of this mechanism. Through functional module analysis, prompt engineering, and behavioral experiments, we provide empirical evidence supporting the counting hypothesis and uncover how models leverage example frequencies to perform inference in few-shot settings. Our findings offer a novel perspective on the internal workings of ICL and lay a foundation for enhancing the interpretability and controllability of large language models.

Technology Category

• Cognitive Modeling & Cognitive Systems: Analogy
• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

In-Context Learning (ICL) indicates that large language models (LLMs) pretrained on a massive amount of data can learn specific tasks from input prompts'examples. ICL is notable for two reasons. First, it does not need modification of LLMs'internal structure. Second, it enables LLMs to perform a wide range of tasks/functions with a few examples demonstrating a desirable task. ICL opens up new ways to utilize LLMs in more domains, but its underlying mechanisms still remain poorly understood, making error correction and diagnosis extremely challenging. Thus, it is imperative that we better understand the limitations of ICL and how exactly LLMs support ICL. Inspired by ICL properties and LLMs'functional modules, we propose 1the counting hypothesis'of ICL, which suggests that LLMs'encoding strategy may underlie ICL, and provide supporting evidence.

Problem

• In-Context Learning
• large language models
• mechanism
• counting hypothesis
• error diagnosis

Innovation

• Counting Hypothesis
• In-Context Learning
• Large Language Models
• Encoding Strategy
• Mechanism Interpretation