Research talk: Unlearning and Relearning in Large Language Models

Title: Unlearning and Relearning in Large Language Models

Speaker: Dr. Fazl Barez

Dr. Barez also spoke on the topic of “Unlearning and Relearning in Large Language Models,” where he explored the importance and challenges of removing undesirable concepts from these models. He emphasized the difficulty in preventing models from relearning such concepts even after they have been fine-tuned. The presentation highlighted the implications for AI safety, underscoring the ongoing challenges and potential solutions for making AI systems more transparent, controllable, and aligned with human values. The session concluded with a discussion on potential NTU-Oxford collaboration.