Andrej Karpathy Proposes AI Benchmark That Builds a 3D Middle-earth
The test extends the 'pelican on a bicycle' benchmark into a more complex 3D generation task, suggesting a new way to track AI's growing ability to build detailed virtual environments.
Reporting from 1 source: GIGAZINE.
OpenAI co-founder Andrej Karpathy proposed a benchmark test that asks AI to generate a 3D world from the opening of The Lord of the Rings. He instructed Claude Opus 5, using three.js, to construct the world with a token limit of one million. The model produced 5,500 lines of code in about two hours, resulting in an animated 3D scene.
Karpathy's test, inspired by Simon Willison's 'pelican on a bicycle' benchmark, asks AI to create a 3D world from the first paragraph of The Lord of the Rings. He fed the text to Claude Opus 5 with a limit of one million tokens. The model spent about two hours generating 5,500 lines of code using three.js, producing an animated scene that recreates the story's opening. The voiceover was generated separately with ElevenLabs, not by Claude.
The output, viewable at karpathy.ai/lotr-movie, shows locations like Bilbo Baggins' hobbit-hole and the cave filled with treasure. While not matching hours of human effort, Karpathy notes the quality is surprisingly high for a two-hour AI-generated piece. He suggests this test, like the pelican benchmark, will likely see rapid quality improvements as AI advances.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.