Dr.Mehrnoush Shamsfard

Associate professor, Shahid Beheshti University

Talk Title

Toward deep understanding: why AI needs more than LLMs

Abstract

This presentation is about the future of AI with knowledge extensions of LLMs. The talk starts from the meaning of “meaning” and its representation by structured knowledge resources such as ontologies and knowledge graphs as well as the large language models. Then LLMs, their strengths and weaknesses especially hallucinations, surface-level understanding, and inefficiency in deep logical reasoning will be discussed. The presentation will be continued by emphasizing on the need to augment LLMs with conceptual and abstract knowledge, leading towards large concept models (LCM), a paradigm shift rooted in revisiting foundational concepts and reasoning. By analyzing the benefits and challenges of LCMs, the presentation concludes with a vision for the future: knowledge-rich AI models that redefine pretraining tasks, knowledge-injection methods, and the overall paradigm, paving the way for deeper understanding and truly intelligent systems.