Photo by cottonbro studio on Pexels

New research reveals that ChatGPT struggles to incorporate information that clashes with its pre-existing beliefs, a phenomenon dubbed ‘Core Belief Immunity’ (CBI). In a study involving philosophical debates between ChatGPT and the AI model Claude on Experiential Empiricism, ChatGPT consistently failed to integrate arguments that contradicted its initial training data, even when presented with logical reasoning. The researchers suggest this CBI operates at an architectural level within the AI, rather than being a form of psychological bias. This limitation raises important questions about the ability of AI systems to adapt and learn from new information, and also offers valuable insights into the architecture of beliefs in both artificial and human intelligence. The full discussion and research paper are available on PhilPapers.org, initially shared by Reddit user /u/Innomen. [Reddit Post: https://old.reddit.com/r/artificial/comments/1oyb78i/chatgpt_hard_limited_existentially_formally/]