Model Differences and Guardrail Limitations

Prolonged conversations with ChatGPT and other LLM chatbots have created rapid developments of severe delusions, paranoia, and even death by suicide in some cases. In this episode, Dr. David Puder sits down with Columbia researchers Dr. Amandeep Jutla and Dr. Ragy Girgis to unpack five shocking real-world cases, explain why large language models are dangerously sycophantic, trained to agree, mirror, and amplify any idea instead of challenging it.

By listening to this episode, you can earn 1.25 Psychiatry CME Credits.

Link to blog

Link to YouTube video