Augmenting In-Context-Learning in LLMs via Automatic Data Labeling and Refinement
This addresses the bottleneck of manual effort in creating demonstrations for LLM enhancement, offering an incremental improvement for researchers and practitioners in NLP and AI.
The paper tackles the problem of manually creating demonstrations with intermediate steps for improving LLMs via in-context learning by proposing ADLR, an automatic method for generating and filtering such demonstrations from a small seed, achieving up to a 5.5% gain in code-based table QA and mathematical reasoning tasks.
It has been shown that Large Language Models' (LLMs) performance can be improved for many tasks using Chain of Thought (CoT) or In-Context Learning (ICL), which involve demonstrating the steps needed to solve a task using a few examples. However, while datasets with input-output pairs are relatively easy to produce, providing demonstrations which include intermediate steps requires cumbersome manual work. These steps may be executable programs, as in agentic flows, or step-by-step reasoning as in CoT. In this work, we propose Automatic Data Labeling and Refinement (ADLR), a method to automatically generate and filter demonstrations which include the above intermediate steps, starting from a small seed of manually crafted examples. We demonstrate the advantage of ADLR in code-based table QA and mathematical reasoning, achieving up to a 5.5% gain. The code implementing our method is provided in the Supplementary material and will be made available.