Always worth remembering that only the finetuned ChatGPT is so filtered. The raw GPT models accessible through OpenAI's API are not, and will happily generate v

Always worth remembering that only the finetuned ChatGPT is so filtered. The raw GPT models accessible through OpenAI's API are not, and will happily generate violence, gore, pornographic content, etc. They never use the phrase "As an AI language model..." - if you ask gpt3.5-turbo "What are you", it will even claim to be a human being. The "AI safety" argument that OpenAI is so fond of is nothing more than cover for policies that avoid making OpenAI look too socially disruptive. The APIs will always be accessed through someone else's frontend so they don't care about making them "safe".