Vector Podcast Code search, Copilot, LLM prompting with empathy and Artifacts with John Berryman
Feb 10, 2025
John Berryman, founder of Arcturus Labs and co-author of "Prompt Engineering for LLMs," shares his journey from search technology to AI. He discusses the evolution of code search and how GitHub Copilot reshapes programming with AI. Berryman emphasizes the balance between automation and human oversight in coding practices. The conversation dives into retrieval-augmented generation (RAG) and the future of interactive digital artifacts, opening discussions on reliability and accountability in AI-generated content. His insights are both informative and thought-provoking.
AI Snips
Chapters
Books
Transcript
Episode notes
Learning by Doing
- John Berryman views his career progression as a series of leaps and bounds, often feeling like "drowning" but ultimately surviving.
- He emphasizes hands-on experimentation as key to understanding LLMs, rather than upfront exhaustive planning.
LLM Unpredictability
- LLM unpredictability requires a shift in how we code, moving away from deterministic expectations.
- Building user experiences that tolerate LLM's finicky nature is crucial for successful applications.
Early Copilot Development
- Early Copilot used completion models with limited context windows (2048 tokens).
- Optimizing context by including relevant code snippets and file paths was crucial for performance.
