Keynote Speaker - Simon Willison
summary
Talk explains how LLMs are practical tools: running Mistral on an iPhone with MLC Chat, using an llm CLI to run local models, and the GPT‑4o/LLama leaderboard vibes. It demos a 6‑minute prototype counter that detects "AI", a Datasette plugin that extracts 40 PSF grant entries into a table, and code interpreter running Python on uploads. Core techniques are retrieval augmented generation and function calling, while the big warning is prompt injection—never mix untrusted text with tool access.
AI-written summary · about
chapters (15)
- 0:00Welcome and introductions
- 3:06Defining artificial intel
- 5:07Understanding LLMs
- 7:15Training data and risks
- 9:24Evaluating model performance
- 12:27Open source and local LLMs
- 15:00Prompt engineering tricks
- 16:48Retrieval augmented generation
- 18:53Function calling and tools
- 20:56Security and injection
- 24:17Code interpreter examples
- 28:44Building side projects
- 32:28Data journalism applications
- 36:07AI ethics and practices
- 40:07Future and conclusion
related talks
12mKeynote: How LLMs Have Changed the Way Organizations Do AI
33mOpening Keynote: LLMs, between fantasy and reality - the poin...- 12mKeynote: How LLMs Have Changed the Way Organizations Do AI
01h 07mKeynote + Conversation
42mKeynote: Designing LLM Native systems
05mKeynote: Teaching LLMs New Knowledge
33mKeynote: Fine-tuning our way towards openness in AI
01h 39mMastering LLMs: From Prompt Engineering to Agentic AI