Epoch After Hours What does economics actually tell us about AGI? – Phil Trammell
62 snips
Oct 1, 2025 
Phil Trammell, a Stanford economist, delves into the complex dance between economic theory and artificial general intelligence (AGI). He critiques existing growth models, suggesting they overlook critical aspects of automation and innovation. The discussion traverses the misrepresentation of GDP during technological transformations and the potential for explosive economic growth due to AGI. Trammell also addresses the misunderstandings surrounding task-based models and offers insights into identifying signs of economic singularities.
AI Snips
Chapters
Books
Transcript
Episode notes
Parallelization Has Hard Limits
- The Jones semi-endogenous growth model overstates parallelizability of R&D and can imply unbounded progress by scaling researchers.
- Phil argues coordination latency and complementarity limit parallel R&D gains, so simple scaling won't guarantee infinite progress.
Coordination Bottlenecks Matter More Than Headcount
- Latency between human brains and coordination costs create diminishing returns to adding isolated compute or researchers.
- Relief of coordination bottlenecks (e.g., larger integrated neural nets) could instead open a new fast growth margin.
Ideas Hardness Varies By Regime
- Historical evidence on 'ideas getting harder to find' is mixed and regime-dependent.
- Phil warns modern R&D differences (profit-driven, new-goods) make naive extrapolation from distant past misleading.
