AMAZINGINDEX.COM 每天向上一点
58.1
VOL. 2026.11
2026.11.14
← 返回 2026.11.14 日报
日报快照 · Daily Snapshot
NO. 019

Perplexity 开源决策模型训练方法

#ARTICLE HackerNews 2026.11.14
推荐指数 65.0 NO. 019 · 2026.11.14
发布2026/11/13Score79Comments17

Perplexity 发布了 Decider v1.3 的完整技术细节,包括训练流程、评估方法和实际应用场景。对于需要构建可靠 AI 决策系统的团队,这是少有的生产级经验公开。

做 AI Agent 的团队普遍卡在'让模型稳定做决策'这一步,之前要么用 ReAct 硬编码流程,要么靠大量 prompt engineering 兜底,但都没解决根本的可靠性问题。Perplexity 把自家决策模型的训练配方全放出来,相当于给了张从'聊天机器人'到'能干活'的路线图。

值得对比的是 Anthropic 的 Constitutional AI 和 OpenAI 的 o1 推理链,Decider 更偏向高频、低延迟的场景决策而非复杂推理。如果你在构建客服自动处理、工单分发、内容审核这类需要快速判断的 Agent 管线,它的评估框架可以直接拿来改。

意见分歧 16 条评论

核心争论:决策模型是否真有创新价值,还是只是云厂商锁定客户的

redanddead

Complete lack of creativity in the field of AI... Here's new product, which is just our shitty version of this easily reproducible technology What is this? Same for Google and their decisions api

jgalt212

Perhaps, but if you're already inside GCP it's nice to have something easy to turn on.

adrianvi

Not just that, if your company approved and gets billed by X cloud, it's quite difficult to request another product from Y cloud

替代方案: Google Decisions APIInception Labs Mercury modelsArtificial Analysis Decision Index
查看原文 →