RSI means faster progress for all models, not a path out of saturation
herbiebradley · x · 2026-10-30
A substantive exchange on AI capability saturation. The counterpart argues frontier models will scale to larger and larger tasks, becoming a "drop-in remote team, remote department, remote company" — noting Claude writing a game already spans many traditional company roles.
herbiebradley counters that this doesn't establish lack of saturation: fundamentally, RSI (recursive self-improvement) implies faster progress for sub-frontier models too. His thought experiment: if you could import Q4 2028 frontier capability at Haiku prices, new larger unsaturated tasks would surely emerge as a frontier revenue source — but can you name a specific thing the average enterprise does today that would remain unsaturated in that scenario? The implication: per-token enterprise demand could be met cheaply, putting the persistence of a frontier premium in doubt.
Related event: Debate Flares Over Whether Frontier Model Token Demand Will Saturate(5 posts)→
More from AGI Musings
- Why you should say "please and thank you" to your AI agents — floguo · 2026-10-30
- Economist on today's job hunt: applicants could never reach a real human — soumitrashukla9 · 2026-10-30
- Cyberattacks aren't a bubble-popping sign — they're AI revenue, argues viral post — inductionheads · 2026-10-30
- Sara Dietschy: Trying to Look Smart Before Models Overtake Us Is Misguided — saranormous · 2026-10-30
- Dev who refused Leetcode for years says AI is the great job-market democratizer — deepakns · 2026-10-30
- Founder swimming against the agent wave says it's 'crashing against a rock' — allanyu_ · 2026-10-30