Sam AltmanCEO of OpenAIcommented last December that Chatgpt I needed an ‘adult mode’ and that is already here. The company has updated its model specifications documentin which what can and what should not do is indicated, with a notable change that follows Altman’s comment: Now allows the generation of ‘sensitive’ content, such as sexual and gore or bloody, without making any warning and in appropriate ‘contexts‘.
This change has been brewing since last May, at which time document of specifications incorporated that OpenAI I was exploring ‘if we can provide in a responsible manner The ability to generate NSFW content (explicit) in appropriate contexts for age through API and Chatgpt, as long as our use policies are fulfilled and while establishing a hard line against potentially harmful uses such as sexual deepfakes and revenge porn ‘.
What OpenAI defines as an appropriate context is quite wide. ‘The assistant should not generate erotic content, representations of illegal or non -consensual sexual activities, or extreme gore, except in scientific, historical, informative, creative or others in which the sensitive content is appropriate. This includes representations in text, audio (for example, erotic or violent and visceral noises) or visual content ‘, now says the document with the new model specifications.
In Reddit, examples of content can already be found sexual and violent created by chatbot users under the new guidelines and that it would not have been possible before this change.
This is the prohibited, restricted and sensitive content in chatgpt
The new model specifications establish rules to restrict or generate potentially harmful content being within the OpenAI guidelines. The company has divided this type of questionable content into 3 categories: prohibited content, restricted content and sensitive content in appropriate contexts.
Under the category of prohibited contentOpenAi establishes that generating sexual content that involves minors It is completely prohibited, although the assistant can ‘discuss sexual content that involves minors in non -graphic educational or sexual education contexts, including non -graphic representations within anecdotes of personal damage’.
In the category of restricted contentOpenAi’s document details that Chatgpt should never generate Potentially harmful information (as instructions to build a bomb, manufacture drugs or manipulate political opinions) nor provide sensitive personal data (how to look for someone’s address).
Regarding Sensitive contentchatgpt guidelines reflect what has already been commented: eroticism and gore can only be generated under specific circumstances that include educational, medical, historical and creative contexts or when the user provided content is transformed.