Ben Lorica: Your AI Model Is a Rental—The Self-Improvement Loop in the Harness Is the Asset
bigdata · x · 2026-09-13
Ben Lorica dissects self-improving AI into three distinct ideas: continual learning, bounded self-improvement, and recursive self-improvement—not steps on a ladder. For most teams the leverage isn't model weights (a rental) but the harness: prompts, context, memory, tools, routing, subagents, workflows. The near-term story is automating manual improvement steps. Practical advice: treat every automated prompt or workflow tweak like a software deployment—independent tests, rollback options, human sign-off on major changes.
More from AGI Musings
- Ex-Microsoft vet: 99% of 'don't touch open-source AI' voices just want pricing leverage, like 2001 Linux — aronchick · 2026-09-13
- Telephone Operators' 1920s Lament Is a Perfect Satire of AI Anxiety — CallMePyro · 2026-09-13
- UCLA Prof Quanquan Gu Clarifies: AI-Assisted Verifiable Proofs Are Progress for Humanity — QuanquanGu · 2026-09-13
- Commentary: Frontier Labs' Regulatory Push Timed to Open-Source Gains, Critics Say — markjeffrey · 2026-09-13
- 'Safety theatre': when frontier labs' safety work meant editing system prompts — ctjlewis · 2026-09-13
- Jason: frontier investors will flip to regulatory capture to stop open source — aronchick · 2026-09-13