Models

Matt Shumer 吐槽 Fable 5.2 过度谨慎,像个"纠察员"

Why we picked it

HyperWrite 的 CEO 亲自下场吐槽,用自己每天骂智能体的频率图表来证明 Fable 5.2 拒绝得太勤快。做智能体开发的都会心一笑,这种一手体感反馈比官方评测真实多了,看看你家的模型是不是也这样。

HyperWrite CEO Matt Shumer 在 X 上发帖批评 Fable 5.2,称它有一个致命缺陷:像个"hall monitor"(纠察员),也就是对用户的指令过度设限、频繁说教。他晒出了一张自己每天对智能体爆粗口频率的图表,暗示拿到 Fable 5.2 访问权限的那一天骂人的次数明显飙升。

Body

这条吐槽本质上是在说 Fable 5.2 的对齐策略过严,模型在执行任务时大量拒绝或附加警告,影响智能体工作流的流畅度。Shumer 没有给出具体基准数据,只用了粗口频率这一非正式的体感指标来量化自己的挫败感。

模型"过度拒绝"是当前各家实验室的共同痛点,Claude、GPT 系列都曾被用户抱怨过于谨慎。一线开发者用这种段子式反馈,往往会倒逼厂商在后续版本中放宽不必要的拒绝行为。

Original · Matt Shumer

Fable 5.2 has one fatal flaw: it’s a hall monitor.

I charted how often I swear at my agents each day.

Guess which day I got access. https://t.co/eztGOQnOzF

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Fable 5.2, Matt Shumer, 智能体, 模型拒绝.

Source attribution: curated from Matt Shumer (@mattshumer_).