Better prompting fixed my LLM-built Flappy Bird: full inputs beat images
maximelabonne · x · 2026-10-06
Maxime Labonne shares hands-on experiments having LLMs build mini-games: his first Flappy Bird clone was a disaster, but with better prompting to provide all necessary inputs, the model became very good. He also tested Quick, Draw! based purely on image inputs against a predefined word list — quite reliable, though not open-ended.
Key takeaways: text inputs already provide everything the model needs (images didn't help), and explicitly feeding complete game state in the prompt is the deciding factor. A follow-up demo with user inputs is planned.
More from coding & agent
- Matt Shumer builds Hogwarts live on Spawn with AI in a multiplayer world — mattshumer_ · 2026-10-06
- Ruff author shares build-speed trick: one crate per file for faster compiles — charliermarsh · 2026-10-06
- Notable MCP critic converts, now wants an official Gmail MCP — zeeg · 2026-10-06
- Developer Runs 1,200+ Blender Modeling Sessions to Blind-Rank LLMs and Agent Harnesses — Izolight · 2026-10-06
- Software-in-the-loop: self-supervised scaling of terminal environments lifts Terminal-Bench 2 to 53.56% — Zhongzhi Li · 2026-10-06
- Real product photos plus LTX-2.5 rotation videos and Codex-coded UI rebuild a brand demo — OpenAIDevs · 2026-10-06