THE HUMAN QUESTION
Why this game was made.
Lumenwake is the game discussed in the author’s Facebook post. The question was simple: what would ChatGPT choose to make if a human supplied the technical brief, then left the gameplay to the model? Successive requests asked it to improve its own work, rather than telling it which mechanics to add. Release notes and design explanations were requested later, after their absence became noticeable.
The author found the result relaxing but somewhat boring, and noticed how firmly the model stayed with its original idea. Procedural generation did not necessarily produce meaningfully different levels. Context compaction also caused some details to be forgotten. Those are candid observations about the original experiment, not automated test results or a claim that every later version is identical.
THE ORIGINAL BRIEF
The prompt.
I want you to imagine a one player game entirely out of your "imagination" - it should run in canvas on a page and should be visually stunning. You have full authority to design the game.
Verbatim initial game request, recovered from the creation conversation.
Then, ask it to improve.
Improve on Lumenwake. As this is completely AI-generated game, I will not steer it in any way; it is up to you to improve the game in a way you see fit.
WHAT THE AI MADE
A small world with its own rules.
Gather light and wake three sleeping stars in a luminous garden. Pulse to collect and repel, dash to evade, rescue echoes and explore seeded or daily gardens.
The human specified the platform and delegated the game design. The current version includes later AI-chosen iterations. Functional checks and the author’s subjective reaction are different kinds of evidence.
Game documentation and release guidance →Inside the game, “Only human intervention” opens its dated release diary. Older rationale is labeled Retrospective interpretation; from v1.6.1, Design decisions records reasons during revisions.