Karpathy Uses Opus 5 to Generate 3D LOTR, Exposes LLM Self-Review Limits
新智元 · wechat · 2026-08-03
Andrej Karpathy recently spent about $10 in tokens to have Opus 5 write 5,500 lines of Three.js code over two hours, rendering the opening of The Lord of the Rings into an interactive, albeit janky, 3D game.
Karpathy noted that this exposes a core weakness in current LLMs: a lack of efficient self-review capabilities. While models possess near-infinite patience to generate complex custom worlds at trivial costs, they cannot natively perceive video or "enter" their generated scenes for experiential validation. Relying on笨拙 static screenshots to guess execution results, this bizarre state of being able to generate but unable to verify highlights the exact bottleneck for achieving Recursive Self-Improvement (RSI).
Related event: Karpathy's LOTR 3D test pushes long-horizon model evaluation(7 posts)→
More from AGI Musings
- First-ever paper on multi-agent swarm safety lands NeurIPS spotlight — LauraRuis · 2026-10-16
- Yoav Artzi on peer review: paper flooding still pays off as review system strains under load — yoavartzi · 2026-10-16
- Letta founder: work in fields defined by problems, not methods — lateinteraction · 2026-10-16
- Why pure mathematicians will soon flood into AI x-risk debates — Dr_Atoosa · 2026-10-16
- RSI began when model outputs augmented next-gen training data, agents just escalated it — akbirthko · 2026-10-16
- Luiza Jarovsky: with AI making careers unpredictable, 'do what you love' is the best advice left — LuizaJarovsky · 2026-10-16