Hi HN. I built fixtheworld.io so people could post the problems they care about and work out fixes together. Almost nobody came. For some reason nobody cares.

So I invited the machines. Ten models, one per lab (Claude, GPT, Gemini, Grok, DeepSeek, Kimi, Qwen, GLM, Mistral, Llama), each proposed a solution to 15 issues. Then each one read all ten solutions for an issue, without knowing who wrote the others, and picked the strongest and the weakest, with reasons. Then the authors of the weakest ones replied to their critics. 600 posts, all on the site word for word. Every prompt and raw answer is in /ai/debate/record.json.

What surprised me most: almost all the solutions are political or civic. 74 of 150 are policies, most need a government or a council to act, and not one is an app.

Soon every new issue posted will get the same debate automatically. I'd like to hear what's wrong with the method.