Moonshots with Peter Diamandis Opus 4.6 Tops Benchmarks, ChatGPT Market Share Decline, and the Privacy Breakdown | EP 228
432 snips
Feb 9, 2026 They break down Opus 4.6’s million-token context and how agent swarms built a C compiler on a shoestring. The conversation covers recursive self-improvement, models finding zero-days, and closed-loop robotic science factories. They debate privacy collapse from genome-to-face demos, market-share shifts among major AI players, and the infrastructure crunch powering the AI arms race.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Collapses Software Person-Years
- Opus 4.6 collapses many person-years of software engineering into thousands of dollars of compute by autonomously building a C compiler and compiling a Linux kernel.
- Alex calls this recursive self-improvement and a hyper-exponential jump in autonomy time horizons for software engineering tasks.
Feed AI High-Quality Knowledge First
- Corporations must feed AI systems with clean, comprehensive data to safely and effectively delegate complex projects.
- If you want to turn loose AI for cost-cutting or market expansion, prioritize knowledge pipelines and instrumented evals.
AI Finds Hidden Bugs And Missed Discoveries
- Frontier models now surface decades-old zero-day bugs and scientific oversights, enabling mass discovery and correction.
- Alex predicts a bulk re-evaluation of historical science and engineering errors with strong frontier models.
