How generative AI models work
Generative AI uses computer models that are trained on billions of pieces of information.
How do these computer models work?
Models are trained using information
This information can come from different places, like publicly available sources on the internet or our own products and services.
Models use information to learn about patterns
The models learn about patterns and relationships between these pieces of information.
Models generate new content based on patterns
The models use what they learn to generate new content, like text or images, based on requests you give. Even though we have safeguards in place to help make sure the responses are appropriate, certain AI features may not be suitable for some teens. Some responses may be inaccurate or inappropriate.How are models updated?
Models are initially trained on a set of information from a moment in time. Here are some ways we improve them and keep them up-to-date.
We make frequent small updatesFor example, when you chat with AI, we may use those messages and replies to improve the model. In some features, you can also tell us what's going well and where things can be improved.
We occasionally take the bigger step of retraining modelsWe occasionally retrain models on a new set of information. Models are retrained on billions of pieces of information, tested and put through safety checks.
Deleted information isn’t used for future trainingModels learn about patterns between pieces of information. If you delete a piece of information a model learned from (like a public post), the model won’t change in that moment. However, your information itself will be deleted, and once it’s deleted, it won’t be used for future training.
