AI skeptic predictions are correlated bets that never cost anything, argues 79K-follower researcher
teortaxesTex · x · 2026-10-31
teortaxesTex (79K followers) dissects a common AI-skeptic argument pattern: skeptics make concrete falsifiable predictions, but the moment one fails they immediately invent the next gate and wave at "a bazillion" more — so losing costs them nothing. The key insight: all these bets are correlated.
The quoted thread also argues:
- Model differences come mainly from training data, not intelligence; there's a hard limit on how much data fits without degrading quality across tasks
- Nobody yet knows how to train a general frontier model without losing quality in some areas
- OpenAI's rumored "math miracle" model is probably only good at math and Lean — its Rust coding is likely no better than GPT 5.1 — and outsiders will probably never see it
- A model that genuinely does math + coding + games + computer use at a sane price would be a real step toward general intelligence; for now it's largely a fiscal-feasibility question
More from AGI Musings
- Dual-AI audit of Saez & Zucman's 138 papers finds 53 defects, 72% of figures reproduce — RexDouglass · 2026-10-31
- Reddit Prediction: All AI Assistants Will Eventually Be Equally Boring and Sanitized — Dogbold · 2026-10-31
- Ford went from inventing the assembly line to full deployment in 3 years, cutting car-building time 88% — 2C_ornot2C · 2026-10-31
- Anti-AI to shield tech-illiterate boomers? Ignoring doctors isn't AI's fault — felpix_ · 2026-10-31
- "AI shouldn't turn everyone into programmers — it should boost existing jobs" — AryHHAry · 2026-10-31
- teortaxesTex pushes back on 'LLMs can't disassemble engines' cope in embodied AI debate — teortaxesTex · 2026-10-31