This Day in AI Podcast EP56: We Wrote a Song! Claude Opus is đź‘‘, Gemini 1.5 Pro & Ultra API Experiments
Mar 28, 2024
Claude Opus, a prominent figure in the LMSYS Chatbot Arena, is highlighted in this podcast. Topics include the release of Google Gemini 1.5 Pro API, Amazon's investments in Anthropic, Hume AI's empathic abilities, and Microsoft absorbing Inflection AI. The discussion also covers the challenges of AI assisting students, stable AI code instruct models, and AI-generated music composition.
AI Snips
Chapters
Transcript
Episode notes
Claude Opus Beats GPT‑4 In Practical Use
- Claude 3 Opus overtook GPT‑4 on the LMCS blind leaderboard for instruction‑following and usefulness.
- Users prefer Opus for consistent, practical assistance and lower refusal/laziness rates.
Wrap Fast Models As Generic Helpers
- Use cheap, fast models like Claude Haiku for routine extraction and formatting tasks as a generic helper.
- Wrap that model call in a small function so your app adapts to variable input without brittle parsing code.
Coding Tests: Sink Sub Showdown
- Chris prompted models to build a Pygame “Sink Sub” game to test code output capabilities.
- Gemini 1.5 produced playable but buggy code after iterations; Claude Opus produced a more polished single‑prompt implementation.