According to Dynamic Beating monitoring, Andrej Karpathy handed the opening paragraph of "The Lord of the Rings" to Opus 5, offering 1 million Tokens and a budget of around $10 to create a 3D world using Three.js.
The model took about 2 hours to run and resulted in 5500 lines of code. It required arranging various polygonal assets in 3D space and then writing code to animate the entire scene. The final product was somewhat rough but did succeed in turning text into a playable 3D scene.
In the past, few were willing to create such content due to the cost and time involved. However, Large Language Models (LLMs) can keep going, turning highly customized worlds into on-the-fly generated content. Karpathy envisions a future where users can enter these dynamically generated worlds and become story characters, spectators, or NPCs, resembling an on-demand "GTA."
The game also exposed the current limitations of LLMs. Opus 5 cannot efficiently or natively watch videos, nor can it playtest games; it can only slowly capture screenshots at different points for inspection. Throughout the process, there were multiple errors, resulting in a substantial amount of rough and faulty content.
While it can already create a world, it is still unable to smoothly review and experience the world it has created.
