Overview
Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a groundbreaking approach to AI personality alignment through latent space manipulation in this 37-minute video presentation. Discover how orthogonal subspaces in artificial intelligence can encode personality traits and complex behaviors by strategically placing additional attention heads on transformer architectures. Learn about the evolution of Large Language Models from general-purpose reasoning systems to specialized, coherent agents capable of maintaining stable psychological profiles for applications ranging from immersive role-playing in open-world environments to empathetic engagement in therapeutic settings. Examine the persistent challenge of achieving personality alignment without compromising the model's core intelligence capabilities. Delve into cutting-edge research on the geometry of persona and techniques for disentangling personality from reasoning in large language models, based on work from the Precision and Intelligence Medical Imaging Lab at Beijing Friendship Hospital, Capital Medical University. Understand how this innovative approach to latent space surgery could potentially revolutionize AI personalization and eliminate the need for traditional fine-tuning methods.
Syllabus
AI Latent Space Surgery: The End of Fine-Tuning?
Taught by
Discover AI