Yoav Goldberg: The Jevons Debate Is Really About LLMs Replacing Data Scientists
yoavgo · x · 2026-09-23
AI researcher Yoav Goldberg weighs in on the Jevons paradox discourse, arguing the real story isn't the economics of a product category but the reaction of people who took over jobs previously done by data scientists, now accomplished with LLMs.
He agrees that making tasks easy (0-shot, ICL), fast, and cheap is genuine value add. The thread also speculates on the RLCD training method: Goldberg guesses it's largely a combination of Verified Rewards and Execution Feedback (or cross-entropy for probabilistic outputs), and asks whether there's any secret sauce beyond that.
Related event: Researchers Debate Jevons Paradox and the Essence of RLCD(4 posts)→
More from AGI Musings
- Rob Leclerc: despite pre-launch alignment drama, new models reach hundreds of millions with no major issues — robleclerc · 2026-09-23
- AI floods French administrative courts as case filings surge 44% in two years — IgorCarron · 2026-09-23
- A UChicago professor used AI to help produce 200+ papers this year — soumitrashukla9 · 2026-09-23
- Alibaba CEO Wu Yongming: machine thinking could hit 1000x human levels; vibe coding is just the lightbulb — prdeepakbabu · 2026-09-23
- Polymarket puts only 19% odds on OpenAI declaring AGI before 2027 despite GPT-6 Astra — Polymarket · 2026-09-23
- Jeffrey Katzenberg on AI creativity: a demo that felt like Luxo Jr. in 1986 — elonmusk · 2026-09-23