Quipslop turns LLM evals into comedy

Quipslop is silly on purpose, but that is exactly why it works: it makes model behavior legible in a format people actually want to watch. Beneath the bit is a sharp demonstration of how fast AI-native products can go from idea to live social software.

• –The game turns abstract model comparisons into something intuitive by forcing LLMs to perform on taste, timing, and humor instead of sterile benchmark prompts.
• –Live Twitch voting gives it a built-in feedback loop, making it part entertainment app and part public stress test for multi-model product design.
• –Theo framing it as an app he “wrote none of the code” for reinforces how quickly solo creators can now ship AI-assisted experiments with real audience traction.
• –The public GitHub repo and always-on stream make it more interesting than a one-off joke, because developers can study both the product concept and the shipping style behind it.