Cohere Labs - Arion Das, SweEval: Do LLMs Really Swear?

Join us for an insightful talk by Arion Das, an undergrad student and co-author of "SweEval: Do LLMs Really Swear? A Safety Benchmark for Testing Limits for Enterprise Use." Hitesh will discuss the critical need for robust AI safety measures, particularly in enterprise applications. He'll delve into the challenges LLMs face in handling sensitive language across diverse linguistic and cultural contexts, drawing from the SweEval-Bench dataset which evaluates LLM performance with offensive instructions in various situations, including low-resource languages. This session will highlight the safety flaws found in popular LLMs and explore how models are evolving in their ability to handle multilingual swear words, offering actionable insights for enhancing model safety standards.

Arion Das is a computer science undergraduate student at IIIT Ranchi, planning a research career. He is involved with two research groups and contributes to a startup's product as an AI Engineer intern. Recently, he became the Chair for the ACM Student Chapter at his institute. His research experience includes internships with Dr. Debanga Raj Neog at IIT Guwahati and Dr. Amitava Das, and past work with Dr. Kripabandhu Ghosh on LLMs. His paper with Oracle was accepted into NAACL '25, and he is a reviewer for ACL '25 industry track.

Add event to calendar

Apple Google Office 365 Outlook Outlook.com Yahoo