Surfacing Biases in Large Language Models using Contrastive Input Decoding
This addresses the challenge of evaluating fairness and robustness in language models for researchers and practitioners, though it is incremental as it builds on existing decoding methods.
The paper tackled the problem of detecting subtle biases in large language models during open-text generation by proposing Contrastive Input Decoding (CID), a decoding algorithm that generates text likely under one input but unlikely under another, resulting in interpretable outputs that reveal context-specific biases.
Ensuring that large language models (LMs) are fair, robust and useful requires an understanding of how different modifications to their inputs impact the model's behaviour. In the context of open-text generation tasks, however, such an evaluation is not trivial. For example, when introducing a model with an input text and a perturbed, "contrastive" version of it, meaningful differences in the next-token predictions may not be revealed with standard decoding strategies. With this motivation in mind, we propose Contrastive Input Decoding (CID): a decoding algorithm to generate text given two inputs, where the generated text is likely given one input but unlikely given the other. In this way, the contrastive generations can highlight potentially subtle differences in how the LM output differs for the two inputs in a simple and interpretable manner. We use CID to highlight context-specific biases that are hard to detect with standard decoding strategies and quantify the effect of different input perturbations.