RSI Paper Sparks Debate: omarsar0 Shares L1-L5 Autonomy Scale to Define Recursive Self-Improvement
omarsar0 · x · 2026-11-18
- AI researcher omarsar0 (Elvis Saravia) shared a table from an RSI survey paper amid a dispute over whether a Google paper truly achieves recursive self-improvement.
- The table (Table 2) organizes representative RSI techniques across autonomy levels L1 (Execution) to L5 (Meta), covering search & optimization (evolutionary search, MCTS, LLM reflection/critique), data & experience construction (self-play, adaptive curriculum), and training & persistent update (SFT, DPO, prompt optimization, memory consolidation), citing works like GEPA, VOYAGER, AZR, and ReasoningBank.
- The core dispute: Doomlaser argues that keeping weights frozen while only optimizing the harness is not RSI and the label is attention-grabbing BS; omarsar0 says he only wants shared terminology to navigate the literature, not to defend the paper.
More from AGI Musings
- Don't force human-like traits on machines—or deny ones they actually have — robertskmiles · 2026-11-18
- Dario on open-weights risk; same model costs 37x more on WorkBuddy vs Doubao — kevinnbass · 2026-11-18
- Are open-source models just linear combinations of Claude and GPT mind styles? — Zergylord · 2026-11-18
- Bank CEOs and NSA tell Congress AI-found vulnerability patches are behind nationwide outages — peterwildeford · 2026-11-18
- HUMAN.md: A Working Manifesto on What Humans Still Bring When AI Averages Everything — Dr_Atoosa · 2026-11-18
- Reddit asks: Hinton's predictions from three years ago still hold — what do big labs think? — SuburbanDad18 · 2026-11-18