The idea proposed by A16Z to use large models as referees for prediction markets is really interesting. As soon as I saw it yesterday, I immediately shared it with a few founders working on prediction markets. Today, I suddenly thought, since the decision-making Prompt is set from the start and locked on-chain, wouldn’t it be more accurate and less controversial to form a committee with multiple large models? GPT + Grok + Gemini + Claude + Deepseek The 'Big Five' of LLMs, decision >= 3/5 votes wins.