The Attention Mechanism with Andrew Mayne Heated Rivalry
11 snips
Mar 31, 2026 Andrew Mayne, former OpenAI science communicator and AI commentator, breaks down the Anthropic vs OpenAI rivalry. He unpacks rumors about Mythos and Spud. He explores model class shifts, agentic tools and brittleness. He discusses compute constraints, Claude’s growing pains, Sora’s cancellation, and how IPO pressure shapes strategy.
AI Snips
Chapters
Books
Transcript
Episode notes
Big AI Step Changes Often Seem Small
- Step changes in AI often feel incremental to consumers despite major technical shifts.
- Andrew Mayne explains recursive reasoning and model-efficiency gains (e.g., chain-of-thought and reasoning models outperforming large training runs).
Agentic Demos Still Too Brittle For Daily Use
- Agentic demos impress but are often too slow and brittle for real users today.
- Andrew and Justin cite Operator and Anthropic's cloud computer demos that worked in demos but failed real-world reliability and speed tests.
Focus On Where You Can Charge A Premium
- Invest where your strengths and customers meet: Anthropic may double down on building a very high-end model.
- Andrew notes Anthropic's history of strong pretraining and code models (Opus) and a market willing to pay for top-tier capability.
