The AI Bubble is FAKE But Getting Fooled Will Cost Your Career—Here’s Why + a Bonus Skills Prompt Pack

Source: Nate’s Newsletter Date: 2025-10-30 URL: https://natesnewsletter.substack.com/p/new-proof-ai-is-exploding-exponentiallybut

Summary

Nate argues that AI capability is genuinely compounding on metrics that can’t be gamed, citing Julian Schrittwieser’s research on exponential improvement curves, and that the “bubble” narrative is a category error—confusing hype cycles in specific applications with the underlying capability curve. The career warning is that the window to build fluency is closing: Nate forecasts a 12–18 month acceleration in economically relevant capability and frames the question as preparation timing, not whether AI matters. The piece includes a 20–25 minute “AI Exponential Fluency” self-assessment that outputs four compounding skills to develop: AI Direction, AI Evaluation, Task Decomposition, and Learning Velocity.

Implications

• Feeds the token economics thread directly: the “bubble is fake, capability is real” argument is a counterweight to the OpenAI subscription collapse data (ChatGPT Plus projected 80% subscriber decline) that lands in April 2026—distinguishing capability growth from business model durability matters for reading that data correctly.
• The four compounding skills (Direction, Evaluation, Decomposition, Learning Velocity) map cleanly onto Nate’s Five Durable Layers: Direction → distribution, Evaluation → trust and taste, Decomposition → context, Learning Velocity → the meta-layer that makes the others sustainable.
• Published October 30, 2025—six months before the “moment of danger” framing (Amodei, May 2026) and the “preparation window is closing” language becomes institutional rather than newsletter rhetoric. Track Schrittwieser’s exponential metrics against actual SWE-Bench and GPQA curves through 2026.