RESEARCHMonitorNEXT 12 MONTHS
From a River in Gilead to the Inference Distributions of Large Language Models: Covert Dialect Bias and Linguistic Profiling at Scale
arXiv cs.CL — Computation and Language
Factual evidence
What the source reports
arXiv research shows LLM alignment fails to eliminate covert dialect bias in internal probability distributions for high-stakes decisions.