Retrieval Augmented Generation: A Technique for AI Models

Retrieval augmented generation (RAG) is a technique where an AI model is applied only to a specific set of relevant and good data, making the model less likely to provide incorrect or new answers. This method is particularly beneficial in business contexts, allowing models to access only internal data sources like a company's internal wiki. While challenging to implement for various daily computer tasks, RAG is a significant step forward. However, the more substantial challenge lies in obtaining diverse data inputs beyond screenshots and audio, especially from activities outside interacting with screens. Accessing a wide range of human experiences, biometrics, and sensory inputs is crucial for enhancing AI tools, although current limitations make it primarily business-focused. These tools aim to assist in recalling and summarizing finite events, such as meetings, rather than encompassing all aspects of daily life and memory. The focus now is on determining the minimum data required for these tools to be effective, like Apple's feature that organizes existing pictures rather than capturing new ones automatically. The key question is how these tools can incrementally provide more utility by enhancing existing data inputs.