LLMs: A Research Perspective

Abstract

The talk explores the current state of the art in large language models (LLMs) from a research perspective. It presents the current limitations of these models such as their lack of compositionality, inability to adequately understand the physical world, limited ability to reason and plan and their tendency to hallucinate. It shows how cutting edge research is addressing such limitations, improving LLMs with a view to making them usefully and safely applicable in more context – and how KInIT is contributing to these efforts.