Here we see host Paul Cradle presenting a mind-blowing almost whistleblower-like account of how he was able to get ChatGPT to tell him how to synthesize Schedule I Controlled substances, something that violates OpenAI's own policies. From there he goes on to extrapolate that because ChatGPT is not even aware of what it's doing enough to stop itself from violating its own policies, while copying verbatim, from its training data, instructions on how to cook up illegal substances, it most certainly will also copy verbatim (or uncomfortably close to verbatim) materials from its own training data, without having any awareness of what it is doing.