Psychology of Modern LLMs

In the video ‘Integrated AI: The psychology of modern LLMs (2024’ by Dr Alan D. Thompson, the presenter explores the concept of ‘integrated AI’ and the psychological aspects of modern large language models (LLMs). Dr. Thompson discusses how LLMs, like humans, develop their own internal representations or ‘world models’ based on the data they are trained on. These world models enable LLMs to generalize, reason, and create, moving beyond simple word prediction to constructing their own understanding of reality. The video highlights the emergence of ‘self models’ in LLMs, which allow them to reflect on their own reasoning and decisions, akin to self-awareness and introspection in humans. Dr. Thompson also introduces the concept of ‘user models,’ where LLMs tailor their responses based on their understanding of the person they are interacting with. This ability to adjust behavior based on user characteristics is compared to human perceptions and biases. The video concludes by emphasizing that the development of these internal models represents a significant shift in AI, moving towards genuine understanding and intellectual partnership with humans. Dr. Thompson’s insights are based on his extensive research and the latest findings in the field, including studies by Harvard and Google.