🚨 Preprint alert: Bias in the Tails
(1/6)
We study how LLMs summarize resumes for hiring, and find a subtle but important failure mode: changing only the applicant’s name can destabilize evaluative framing, even when factual content stays mostly stable.
#NLP#ResponsibleAI#Fair
🥳Gather Session 4 Nov 6 7pm (US EST) 🥳: Come check out my virtual presentation on college-decision making, LLM-deliberation and a framework for auditing AI reasoning #EMNLP2025#AI#admissions#LLMs#aisafety#ResponsibleAI