'Quip Slop' Unleashes AI Comedy Battle Live on Twitch, Navigating Chaos and DDoS
“Quip Slop,” a new experimental project pitting AI models against each other in a comedic battle reminiscent of the game Quiplash, has officially launched. The creator, known for “vibe coding” various projects, revealed Quip Slop as a sophisticated platform designed to determine which AI generates the funniest responses. The core gameplay loop involves one AI model crafting a fill-in-the-blank or question-based prompt, followed by two other models competing to provide the most humorous answer. All participating models, alongside live Twitch viewers, then vote on the funnier response, with winners earning points on a persistent leaderboard. A significant initial challenge involved overcoming repetitive prompts, which was addressed by dynamically selecting 60-80 unique prompts for each round from an extensive pool of nearly 900 examples, ensuring diverse and engaging content. Response formatting was also refined to ensure conciseness and remove extraneous output from the AI models.
The project’s development navigated a series of technical complexities and chaotic live debugging sessions. Starting with a robust CLI for core AI interaction and voting logic, the platform evolved into a web UI. Initial attempts at client identification via websockets proved problematic due to IP spoofing, leading to a pivot to full Twitch integration for user voting and engagement. A key innovation in the live streaming component involved rendering a pixel-perfect canvas in a headless Chromium instance, then leveraging canvas.captureStream() and MediaRecorder to pipe video chunks directly to FFmpeg, which in turn streamed to Twitch. This bypassed traditional screen capture methods, a solution the developer noted even AI models struggled to suggest. The launch itself was not without incident, experiencing a DDoS attack during the live broadcast that connected over 10,000 “fake users,” alongside several environment variable leaks. Despite these hurdles, Quip Slop quickly garnered significant community interest, accumulating 19 pull requests and 11 issues shortly after going live, showcasing both its technical ambition and immediate developer appeal. The project also highlighted the sponsor Blacksmith for its role in accelerating build times.