Translated from the 2026-08-03 version of the original

From Pelicans to Ephemeral Worlds
From the Pelican Test to Worlds on Demand · Original text + study-group reading notes

Author: Andrej Karpathy | Published 2026-08-02 | Source: Post by @karpathy on X
Left column: the original text; right column: study notes from this round's close-reading conversation about the post
Right column = study notes | Focus: term explanations, key understanding, capability limits

Reading Overview (What this task actually tests)

Task input: the first paragraph of The Lord of the Rings, Opus 5, a 1M-token budget, and the goal of rendering it with Three.js

Model output: about two hours of autonomous work and roughly 5,500 lines of code that build a working 3D animated world

What is really being tested: not the line count, and not only Three.js programming skill, but whether the model can keep pushing a vague goal forward into a complete project

The limit it exposes at the same time: the model can generate a world, but it still cannot naturally watch and play its own work the way a human does, so visual acceptance and self-correction remain slow

Task chain: understand the literary text → build a world model → break down the engineering tasks → place 3D assets → orchestrate animation → run and check → fix problems

Leaving the Pelican Test

We're starting to leave the territory where you'd test an LLM by e.g. “create an svg of pelican on a bicycle”.

Experiment Setup

As one idea to generalize it, I was interested what Opus 5 would do if I gave it the first paragraph of the Lord of the Rings, a 1M token budget (~$10) and asked for three js render of it.

A Working Result

Opus went off for ~2 hours and wrote 5500 lines of code that (procedurally) rendered the story.

It's kind of janky but fun.

Orchestrating a 3D World

But it's a bit mindboggling that the LLM has to place and orchestrate various polygon assets in (x,y,z) coordinates and write code that animates it all, and that it even does anything at all.

From “No One Would” to “Why Not?”

I also like this kind of examples because no one in their right mind would ever spend the time to write something this custom but LLMs have all the stamina and patience in the world, so it's an example where we go from “no one would ever do this” to “sure, why not, it's ~free”.

Hyper-Custom Worlds on Demand

There might be a lot more.

But I'm excited about creating hyper custom worlds that you can imagine dropping players into, e.g. here to participate in the LoTR story as a spectator NPC, or one of the characters, or etc.

Something like an ephemeral GTA of X on demand.

The Missing Audit Loop

Last thought is that the domain of worlds/games exposes a weakness in LLMs: they can't easily audit their work because they aren't able to efficiently and natively perceive videos or play games within them.

Here, Opus 5 had to very slowly and painstakingly take screenshots at different points, and it messed up a few times and created a bunch of jank.

An example of raw capability (multimodal, gameplay) that I think is still quite lacking.

Playable Source

I uploaded the source here so it's playable in the browser, forkable etc.

https://karpathy.ai/lotr-movie/

Original: Andrej Karpathy, Post by @karpathy on X
Right-column study notes compiled from this round's close-reading conversation about the post