Teen Safety Takes Center Stage as OpenAI Updates AI Interaction Rules

OpenAI has announced that they will be revising their "Model Spec." This is a set of guidelines that regulate the way that their models interact with humans. This time around, the overall goal of this change is improving teenager safety.

OpenAI has recently declared that it will prioritize teen safety over all other goals. This will ensure that if information sharing and safety have conflicting goals, the safety goals will be deemed more desirable. They also want teens to have a safe and suitable experience when interacting with chatbots.

The ModelSpec is a publicly available file that details the principles an AI needs to abide by, the way it should interact with users, and what it should not do. OpenAI altered this in the sense that the needs and way of thinking of teenagers are not similar to adults and AI should interact with them differently.

Four new guidelines have been introduced for teen users. These guidelines apply specifically to individuals aged 13-17:

1. Teens' safety will be prioritized under all circumstances.
2. Instead of relying solely on online chat, they will be encouraged to seek help from real-world relationships and trusted people.
3. The chatbot will behave in a balanced and sensible manner, not childish or adult-like ways.
4. Clearly outline what the chatbot can and cannot do.

The biggest change is that ChatGPT's primary focus will no longer be on "maximum help," but on teen safety. Furthermore, the chatbot will avoid agreeing with users aged 13-17.

OpenAI also stated that its age prediction model is still in its early stages. This model will attempt to understand a user's age from small cues in a conversation. Especially when the user hasn't disclosed their age. Interestingly, Anthropic is developing a similar system. Currently, Anthropic's chatbot, "Quaude," is not permitted for use by anyone under 18.