HOW DOES CHATGPT WORK? AS EXPLAINED BY THE CHATGPT TEAM

Summary from the TLDR newsletter

When users ask ChatGPT a question, OpenAI chunks the question into tokens, creates embeddings, multiplies the embeddings by model weights, and then samples a prediction. This article explains this process in detail. It covers pretraining and inference as well as the system's limitations. Links to other resources that also explain how ChatGPT works are available in the article.