Learn About Embeddings
The chat window on the home page talks to OpenAI's Chat GPT-4 model directly, using a prompt we have crafted over the course of many conversations with Sati. The idea there is that you get to talk with all the information GPT-4 was ever trained on, through the specific and carefully crafted lens of our prompt, which ensures that Sati responds with particular references and in a curated style.
There are other ways of adapting LLMs to meet our educational and other needs. On this page, we're going to be using “embeddings”, which is a method by which we can train an LLM on very specific input data - books, articles, videos etc. - that are especially relevant to the context of mindfulness and meditation within the Theravada tradition of Buddhism, thereby making its response even more particular.
These representations capture the essential features of the original content, which allows Sati to understand and converse with you about it. It is eerily like talking with the authors themselves. You can make your own embeddings very easily by following the link below:
Some of the sources for the embeddings we used to train this version of Sati are listed below. They include books, video interviews, and a collection of different articles.