A Trip Towards Fairness: Bias and De-Biasing in Large Language Models
This addresses fairness issues in democratized large language models, which could harm downstream NLP applications, though it is incremental as it applies existing debiasing methods to new models.
The paper investigated bias in three families of cheap-to-build large language models (CtB-LLMs), finding that bias depends on perplexity rather than parameter count, and demonstrated that debiasing techniques like LoRA can reduce bias by up to 4.12 points in normalized stereotype scores.
Cheap-to-Build Very Large-Language Models (CtB-LLMs) with affordable training are emerging as the next big revolution in natural language processing and understanding. These CtB-LLMs are democratizing access to trainable Very Large-Language Models (VLLMs) and, thus, may represent the building blocks of many NLP systems solving downstream tasks. Hence, a little or a large bias in CtB-LLMs may cause huge harm. In this paper, we performed a large investigation of the bias of three families of CtB-LLMs, and we showed that debiasing techniques are effective and usable. Indeed, according to current tests, the LLaMA and the OPT families have an important bias in gender, race, religion, and profession. In contrast to the analysis for other LLMs, we discovered that bias depends not on the number of parameters but on the perplexity. Finally, the debiasing of OPT using LoRA reduces bias up to 4.12 points in the normalized stereotype score.