Robust and Efficient Frontier Pipelines for Complex Knowledge-Intensive Tasks in the LLM Era

Join a research talk exploring the challenges and solutions in Complex Knowledge Intensive Tasks (CKIT) presented by Dr. Venktesh Viswanathan from TU Delft. Discover how modern pipelines handle complex fact-checking and question answering through multistep reasoning using hybrid knowledge sources. Learn about the current limitations of Large Language Models in CKIT applications and explore innovative solutions including exemplar selection methods with theoretical guarantees and neighborhood-aware retrieval approaches. Understand the practical applications of these technologies in healthcare, where they assist multidisciplinary teams in protocol decisions, and in journalism, where they streamline research processes. Gain insights into the latest developments in retrieval and reasoning capabilities, their limitations, and future directions for advancing robust frontier pipelines in the era of Large Language Models.