Kevin Bryan Says Minimal RSI Could Arrive Around 2027 as AI R&D Productivity Rises 9%
imjustnewatai · x · 2026-07-29
Kevin Bryan says multiple high-level researchers expect a “minimal RSI” breakthrough around 2027: an AI produces a Chinchilla-law-level insight, tests it, and helps shape the next training run.
A separate METR/Elasticity paper estimates that since coding agents arrived, each one-unit increase in model capability has lifted AI R&D productivity by about 9%. Their back-of-the-envelope threshold for self-sustaining acceleration is around 15%.
The implication is that a model-authored frontier breakthrough could arrive before full self-sustaining takeoff. The key number to watch after 2027 is whether each capability step yields 9%, 12% or 16% more research output.
Related event: AI Researchers Predict Minimal RSI Breakthrough Around 2027(2 posts)→
More from AGI Musings
- Anthropic backs petition to pace frontier AI development after recursive self-improvement study — basedjensen · 2026-07-29
- A long rant says quantum AI is about ontology, not speed — AryHHAry · 2026-07-29
- AI closes 90% of support tickets, exposing culture as the new bottleneck — DanielLockyer · 2026-07-29
- University students debate whether AI-written essays can really be detected — fairywitchfr · 2026-07-29
- AI mania may peak in 2028, with a bubble burst in 2029 and decades of growth after — sujingshen · 2026-07-29
- Anti-AI-pause post says slowing frontier progress is “anti-human” — basedjensen · 2026-07-29