展示HN：Maestro – 一个用于协调和整合竞争AI模型的框架
我花了几个月的时间设计一个框架，用于并行协调多个大型语言模型——不是为了选择“最佳”模型，而是让它们进行争论、混合输出，并在结构上保留不同意见。
这个框架叫做Maestro，以下是白皮书链接：
<a href="https://github.com/d3fq0n1/maestro-orchestrator">https://github.com/d3fq0n1/maestro-orchestrator</a>
（叙述版本在这里：<a href="https://defqon1.substack.com/p/maestro-a-framework-for-coherent" rel="nofollow">https://defqon1.substack.com/p/maestro-a-framework-for-coher...</a>）
核心思想：
- 提示被分发到多个大型语言模型（例如，GPT-4、Claude、开源模型）
- 系统比较它们的输出并进行综合
- 它从不归结为单一声音——最终遵循66%的规则：2票支持主要输出，1票保留异议
- 人类评论员和类比验证者可以被触发以进行现实世界的确认（当声明需要依据时）
- 反馈循环不仅学习正确/错误的输出，还学习哪种类型的分歧能引导更深层的真理
Maestro不是一个产品或API——它是一个关于合成智能的开放公民层的提案。它旨在确保认知的完整性并抵制集中控制。
欢迎分享想法、批评或合作。
查看原文
ive spent the past few months designing a framework for orchestrating multiple large language models in parallel — not to choose the “best,” but to let them argue, mix their outputs, and preserve dissent structurally.<p>It’s called Maestro heres the whitepaper
<a href="https://github.com/d3fq0n1/maestro-orchestrator">https://github.com/d3fq0n1/maestro-orchestrator</a>
(Narrative version here: <a href="https://defqon1.substack.com/p/maestro-a-framework-for-coherent" rel="nofollow">https://defqon1.substack.com/p/maestro-a-framework-for-coher...</a>)<p>Core ideas:<p>Prompts are dispatched to multiple LLMs (e.g., GPT-4, Claude, open-source models)<p>The system compares their outputs and synthesizes them<p>It never resolves into a single voice — it ends with a 66% rule: 2 votes for a primary output, 1 dissent preserved<p>Human critics and analog verifiers can be triggered for physical-world confirmation (when claims demand grounding)<p>The feedback loop learns not only from right/wrong outputs, but from what kind of disagreements lead to deeper truth<p>Maestro isn’t a product or API — it’s a proposal for an open, civic layer of synthetic intelligence. It’s designed for epistemic integrity and resistance to centralized control.<p>Would love thoughts, critiques, or collaborators.