RSI is Like NAS: Still Bounded by Search Space
kamalgupta09 · x · 2026-07-15
The author compares the research atmosphere around Recursive Self-Improvement (RSI) to NAS (Neural Architecture Search) around 2018: both are constrained by a "pre-defined search space."
He argues that NAS couldn't escape the convolutional network paradigm back then; similarly, while RSI can write new code, alter algorithms, and read papers, it remains limited by:
- External scaffolding
- Human-defined validation loops
The conclusion is that current RSI progress is real, but whether it can truly break through these artificial boundaries and achieve "escape velocity" remains unclear.
More from AGI Musings
- Andrew Blumberg says formalization without interpretability is not science — AlexKontorovich · 2026-07-21
- Ken Ono says AI is forcing mathematicians to rethink how discovery works — soumitrashukla9 · 2026-07-21
- Open-source labs could distill a state-of-the-art model to 32GB or 80GB VRAM, the post argues — bookwormengr · 2026-07-21
- Two US companies are now using superintelligence to speed up the next generation of models — yacineMTB · 2026-07-21
- MIT Sloan says information, national security and finance are most exposed to AI — Exp_Mark · 2026-07-21
- IMF says AI could lift Sub-Saharan Africa’s economy by 4% over the next decade — Polymarket · 2026-07-21