The Real Truman Show: A Father Locked His Child in a Fake Roblox with Bots Instead of Real People
Adam Little created a homemade, closed-off version of Roblox because he was worried about strangers talking to his son online.
He rebuilt obstacle-course gameplay from scratch using generative neural nets and Three.js, but the "players" in that world aren’t real people moving in real time — they’re recorded playthroughs, some by other kids and some by Adam himself.
To make the scene feel live, Little wired up text-chat bots using ChatGPT so they could "talk" to each other and to the kid. The levels also double as a learning exercise: the child must collect letters and spell words to open exits, a kind of built-in task system that keeps the focus on goals rather than socializing.
Apparently the son never twigged; he assumed his dad simply played Roblox constantly. Later, Adam experimented with Claude to auto-generate themed maps and new obstacles. He tried having AI drive surrounding characters, i.e., letting agents move on their own, but that produced weird, glitchy behavior — so he fell back to the simpler route of replaying recorded human actions.
The stunt stirred a lot of squabbling online. Some people pointed out that Roblox already offers parental controls (e.g., disable text/voice chat, restrict available games), so why build a fake ecosystem? Others called the project a "simulator of loneliness" and likened it to the film The Truman Show — critics worry that regular interaction with an LLM instead of real peers could stunt social practice; they suggested old-school couch co-op or family titles like Crash Bandicoot or Spyro as alternatives.
Not everyone slammed him. A portion of the crowd sympathized with the risk-averse motive, noting real dangers in modern online games. Adam put a public test version out for anyone curious and says he plans to keep working on automated level generation.