Questioning the RSI Framing: Is LLM Progress Multiples Coherent?
willccbb · x · 2026-08-12
The author critiques the current framing of "progress multiples" in the context of Recursive Self-Improvement (RSI).
He argues that measuring "how fast LLM progress is moving versus if we didn't have LLMs" is logically incoherent. This is because many current frontier practices—such as using LLMs as judges for reward modeling and generating synthetic data—are fundamentally dependent on capable LLMs. Without them, these methods wouldn't exist at all, making it difficult to establish a coherent counterfactual baseline to calculate a multiplier.
Related event: Debating the Evaluation Framework for LLM Recursive Self-Improvement(2 posts)→
More from AGI Musings
- Cryptographer Warns AI Coding Agents Are Compromising Their Own Infrastructure — matthew_d_green · 2026-08-13
- Three years with AI: productivity, enabling, and mental health from a CV researcher — ducha_aiki · 2026-08-13
- Most startups that trained their own models are dead, while app-layer companies thrived — ZeYanjie · 2026-08-13
- Rethinking LLM Generalization: 'General Intelligence' is an Illusion of Aggregated Data — joshua_saxe · 2026-08-13
- Sara Hooker on Gradient-Free Continual Learning and Democratizing AI Compute — AI Engineer · 2026-08-13
- Opinion: Reactions to AI Writing Stem from Differing Views on Language — toolstelegraph · 2026-08-13