kuza55 and Jules Jacobs Debate Whether Stacking S-Curves Can Sustain AI Progress
In an August 29 discussion, kuza55 and Jules Jacobs exchanged views on the sustainability of AI progress. They broadly agreed current methods will keep working for a while, but debated whether S-curves can be stacked indefinitely to sustain long-term progress—an open question that makes the discussion worth following.
Confirmed
- kuza55 believes software engineering problems will clearly be solved, and progress is currently fast; but outside software/computer use/math, the trajectory may be weaker
- On how high the current AI S-curve's endpoint sits, Jules Jacobs asked whether it's above "building a better C compiler than GCC/LLVM"; kuza55 answered it would definitely be higher, saying "we're close"
- Jules Jacobs argued there's no need to turn everything into RL environments—just turn AI development itself into an RL environment
- kuza55 raised two potential bottlenecks: compute funding running out and data running out
Unconfirmed
- Both sides admit uncertainty about how long S-curves can keep stacking; Jules Jacobs considers an AI-built compiler better than GCC/LLVM to be something never achieved before
Why it matters
- The discussion makes "can AI keep progressing" concrete and testable: can software engineering be solved, and can a compiler surpassing GCC/LLVM be built—providing yardsticks for tracking future progress
- The idea of "making AI development itself an RL environment" is a potential path around data and compute bottlenecks
2026-08-29 ~ 2026-08-29 · 8 related posts
Primary sources
- S-Curve View: Uncertainty on Stacking AI Paradigms — kuza55 ·
- Are we hitting limits? AI S-curves and scaling uncertainty — JulesJacobs5 ·
- Jules Jacobs: you only need AI development itself as an RL environment — JulesJacobs5 ·
- [source] Jules Jacobs: you only need AI development itself as an RL environment — JulesJacobs5 · 2026-08-29
- [source] S-Curve View: Uncertainty on Stacking AI Paradigms — kuza55 · 2026-08-29
- [source] Are we hitting limits? AI S-curves and scaling uncertainty — JulesJacobs5 · 2026-08-29
- AI coding S-curve peak: Better than GCC/LLVM? — kuza55 · 2026-08-29
- Software engineering will be solved, but AI struggles outside software/math — kuza55 · 2026-08-29
- If software gets solved, could an automated AI factory self-train robotics models? — JulesJacobs5 · 2026-08-29
- Robotics may be stuck inside the LLM S-curve without sim2real or RL efficiency breakthroughs — kuza55 · 2026-08-29
- Robotics breakthroughs need Sim2Real or RL efficiency leaps off the LLM curve — kuza55 · 2026-08-29