Fine-tuning vs. In-context Learning in Large Language Models: A Formal Language Learning Perspective

TL;DR

This paper compares fine-tuning and in-context learning in large language models using a formal language learning framework, revealing differences in proficiency, inductive biases, and model sensitivities.

Contribution

It introduces a formal language learning task and a discriminative proficiency test to rigorously compare FT and ICL modes in LLMs.

Findings

FT outperforms ICL on in-distribution generalization

Both modes perform similarly on out-of-distribution generalization

Inductive biases are similar at partial learning but diverge at higher proficiency

Abstract

Large language models (LLMs) operate in two fundamental learning modes - fine-tuning (FT) and in-context learning (ICL) - raising key questions about which mode yields greater language proficiency and whether they differ in their inductive biases. Prior studies comparing FT and ICL have yielded mixed and inconclusive results due to inconsistent experimental setups. To enable a rigorous comparison, we propose a formal language learning task - offering precise language boundaries, controlled string sampling, and no data contamination - and introduce a discriminative test for language proficiency, where an LLM succeeds if it assigns higher generation probability to in-language strings than to out-of-language strings. Empirically, we find that: (a) FT has greater language proficiency than ICL on in-distribution generalization, but both perform equally well on out-of-distribution generalization. (b) Their inductive biases, measured by the correlation in string generation probabilities, are similar when both modes partially learn the language but diverge at higher proficiency levels. (c) Unlike FT, ICL performance differs substantially across models of varying sizes and families and is sensitive to the token vocabulary of the language. Thus, our work demonstrates the promise of formal languages as a controlled testbed for evaluating LLMs, behaviors that are difficult to isolate in natural language datasets. Our source code is available at https://github.com/bishwamittra/formallm.