Ex-Grok contributor built a Twitter comedy data backlog to train a comedy engine
ivan_bezdomny · x · 2026-11-03
ivanbezdomny discusses his comedy-generation project with @spakhm: he has saved years of Twitter comedy as potential training data, though he's starting with other data first. His thesis: if a model can't do roasts, a comedy engine can't do anything more complicated. He tried a similar idea in the early Grok days; today's models are much better at gathering and organizing training data — but still not at comedy.
Related event: Developer builds Claude-powered AI comedy engine, roast jokes now passable(10 posts)→
More from Fun
- 'SaaS is dead' hot take gets a one-line rebuttal: people cook, restaurants still exist — sanderssays · 2026-11-03
- "Dataset Cartographer": a playful rebranding of the humble data analyst — wavefnx · 2026-11-03
- Palantir CEO Karp: "Once You Go Palantir, You Don't Go Back" — eliano · 2026-11-03
- 'SIX FINGERS': AI girl sings through every era of AI video, 2021 to now — ClaudiaSlop · 2026-11-03
- Meme: My 30-year-old fridge watching the new smart fridges get cyberattacked — evilsocket · 2026-11-03
- Ranking the 400 Most Famous LLMs Ever Made — and the 'Heartbreak' Bottom Tier — teortaxesTex · 2026-11-03