Interacting with LLMs with Minimal Chat

Eugene Yan · 3 years ago

A researcher prototyped a book recommendation interface that minimizes chat-based interaction with large language models by leveraging user context from clicks and browsing behavior instead. The system uses approximate nearest neighbor search on item embeddings to retrieve similar books, pre-cached vibe keywords for filtering, and an LLM librarian for personalized answers, combining recommendation systems with LLM capabilities. This approach suggests that LLM applications could prioritize implicit context signals and direct interaction over chat as the primary mode of engagement.