Simon Willison’s Weblog

The Dual LLM pattern for building AI assistants that can resist prompt injection

I really want an AI assistant: a Large Language Model powered chatbot that can answer questions and perform actions for me based on access to my private data and tools.