Mitigating Hallucinations with Reliable Citations: User and Developer Perspectives
How can you check whether an LLM-generated answer is supported by its sources—and whether its citations actually support the claims?
Large language models can produce text that is unsupported by the materials they are expected to rely on. Citations can help users verify an answer, but identifying errors in both the generated text and its supporting evidence remains a challenge.
Join Professor Ido Dagan, founder of the Natural Language Processing Lab at Bar-Ilan University, for a research-led discussion of hallucinations and reliable citations. We’ll explore approaches that connect generated claims to specific supporting evidence, how to analyze hallucination and citation errors, and why tracing these errors becomes harder in agentic systems.
What you’ll learn
- Verifying LLM output: How localized supporting evidence can help users check generated claims against source material.
- Understanding citation errors: How to identify and analyze errors in both an answer and the evidence presented to support it.
- Tracing errors in agentic systems: What makes more complex workflows harder to inspect, and the challenges this creates for users and developers.
Who should attend
AI and NLP researchers, engineers and practitioners in academia and industry who want a deeper understanding of LLM hallucinations, citations and the verification of generated text