Ben Lorica: practical self-improvement arrives in the AI harness long before RSI
bigdata · x · 2026-09-09
In his Gradient Flow newsletter, Ben Lorica separates three conflated ideas: continual learning (getting better from experience), bounded self-improvement (turning experience into tested, persistent changes), and recursive self-improvement (improving the process that generates improvements). These aren't rungs on a ladder.
Key points:
- RSI matters mostly to frontier labs, but useful automation tools will arrive well before full RSI for most application teams
- The leverage is not in model weights but in the harness: prompts, context, memory, tools, routing, subagents, and workflows — the part teams actually own
- The near-term story is automating pieces of the improvement process currently done by hand: learn from production failures, propose changes to the application stack, test, keep what works
- The model is a rental, but the improvement loop is an asset
More from AGI Musings
- AGI Fellowship kicks off at Johns Hopkins Bloomberg Center — ghadfield · 2026-09-09
- Railroads in the 1830s: the only real historical parallel to AGI — Traditional-Chip8339 · 2026-09-09
- Pedro Domingos: 'AlphaGo didn't beat Lee Sedol — the humans who built it did' — pmddomingos · 2026-09-09
- Mechanize's Tamay Besiroglu: why labs achieve the same breakthroughs at the same time — tamaybes · 2026-09-09
- Connor Leahy: the plan for superintelligence is build it first, figure out control later — victor_explore · 2026-09-09
- Researcher: good vs bad AI futures hinge on DL training's alignment generalization — QuintinPope5 · 2026-09-09