A Study on the Calibration of In-context Learning
This work addresses uncertainty quantification for safe deployment of language models, but it is incremental as it builds on prior calibration research.
The study investigated the calibration of in-context learning (ICL) in language models, finding that miscalibration increases initially with more examples before improving, and that methods like fine-tuning and chain-of-thought prompting can worsen calibration, while a scaling-binning calibrator consistently reduces errors.
Accurate uncertainty quantification is crucial for the safe deployment of machine learning models, and prior research has demonstrated improvements in the calibration of modern language models (LMs). We study in-context learning (ICL), a prevalent method for adapting static LMs through tailored prompts, and examine the balance between performance and calibration across a broad spectrum of natural language understanding and reasoning tasks. Through comprehensive experiments, we observe that, with an increasing number of ICL examples, models initially exhibit increased miscalibration before achieving better calibration and miscalibration tends to arise in low-shot settings. Moreover, we find that methods aimed at improving usability, such as fine-tuning and chain-of-thought (CoT) prompting, can lead to miscalibration and unreliable natural language explanations. Furthermore, we explore recalibration techniques and find that a scaling-binning calibrator can reduce calibration errors consistently.