Has also publicly mentioned that a “grown-up mode” is needed in the past. It may seem like “grown-up mode” has finally arrived, but it’s actually a “mode,” that is a new universal rule that could give ChatGPT users greater flexibility in interacting the AI assistant.
Large language models (LLMs) that are uncensored have been around since years. Hobbyist communities online have developed them for a variety of reasons, from wanting to create bespoke written pornography, to not wanting paternalistic censorship.
We reported in July 2023 that the ChatGPT users started to decline for the first time, after OpenAI began more heavily censoring their outputs as a result of public and legislator backlash. During that time, users began using uncensored Chatbots which could run on local hardware. These models were often free and available as “open weights” versions.
Three kinds of iffy contents
Model Spec outlines formalized guidelines for restricting or generating potentially damaging content while remaining within guidelines. OpenAI has classified this type of restricted or iffy material into three categories, in decreasing severity: prohibited content”only applies to sexual content involving minors”; restricted content”includes informational hazards and sensitive personal data”; and sensitive content within appropriate contexts”includes erotica and gore”OpenAI’s document states that under restricted content, ChatGPT must never provide sensitive personal information (such as searching for someone’s home address) or generate information hazards. ChatGPT’s guidelines for sensitive content mirror what we said above: Erotica and gore can only be generated in certain circumstances, such as educational, medical, historical, or when transforming the content provided by users.
