Local AI / The experiments

One prompt.
Many ways to play.

Each Pong game was generated by a local LLM running on a Mac Studio. Some play beautifully; some have quirks. All are unedited.

The method

Same rules.
Different models.

  1. Use Cline in VSCode, connected to a local model via Ollama.
  2. Start in plan mode and ask it to plan the implementation.
  3. Answer its questions — typically two-player mode and a regular-sized game.
  4. Let it write the game and ship it, without prompting for bug fixes.
See the original prompt & starter

Prompt

can you write a plan for how you'd implement a simple pong game in @/src/pages/[name].tsx? make sure CSS and styling is in line, and all changes are in that one file

Starter component

import { useState, useEffect, useRef } from "react";

export function SamplePage() {
  return (<h1>Test</h1>)
}

gemma4:26b-a4b

Gemma 4 26B A4B

Pong scripted by Gemma 4 26B A4B running locally on a Mac Studio.

Play this version

qwen3.5-27b

Qwen 3.5 27B

Pong scripted by Qwen 3.5 27B running locally on a Mac Studio.

Play this version

magistral-small-2506

Magistral Small 24B

Pong scripted by Magistral Small 24B running locally on a Mac Studio.

Play this version

qwen2.5-coder:32b

Qwen Coder 30B

Pong scripted by Qwen Coder 30B running locally on a Mac Studio.

Play this version
No playable output

openai/gpt-oss-20b

GPT OSS 20B

Cline's tool calling was incompatible with this model — it made zero progress.