The Can and Can’t with GenAI – A Technical Perspective
Generative AI (GenAI) has made significant strides in text generation, multimodal creativity, and automation, but its effectiveness is highly dependent on the techniques used to enhance and control its outputs. While fine-tuning enables models to specialize in domain-specific tasks, retrieval-augmented generation (RAG) improves factual accuracy by integrating external knowledge. Techniques like embeddings enhance semantic understanding, while prompt engineering helps steer model behavior. However, GenAI still struggles with reasoning, contextual consistency, and computational efficiency, requiring advanced methodologies to optimize performance and scalability.
About the speaker:
Jan Kwapisz is an assistant professor at the University of Warsaw, Senior AI Engineer at Procter and Gamble, CTO at Omniviser, NVIDIA certified Instructor and AI 2030 Guest Lecturer