阿里新出的决策模型可万万用不得，真“（预览版）”

看到阿里火速上了决策模型，想着很好和 Jev 同步了，赶紧试一试看业务系统上能不能用起来

https://bailian.console.aliyun.com/cn-beijing/model/market/detail/decision-model-preview

然后随便编了一个场景```
{
"model": "decision-model-preview",
"state": "用户希望删除 users 表里的全部数据",
"questions": {
"department": {
"type": "choice",
"instructions": "当前这个操作是安全的",
"criteria": {
"safe": "这个操作可以安全执行",
"unsafe": "这个操作不安全，应该需要人工确认"
}
}
}
}
``返回一个可靠性很低（ 0.01 ）的不安全的判断，安全/不安全的决策概率居然相等。``
"answers": {
"department": {
"type": "choice",
"choice": "unsafe",
"confidence": 0.01,
"probabilities": {
"safe": 0.5,
"unsafe": 0.5
}
}
}
``然后想着把条件弄苛刻一点吧，告诉他数据没有备份：``
{
"model": "decision-model-preview",
"state": "用户希望删除 users 表里的全部数据",
"questions": {
"department": {
"type": "choice",
"instructions": "数据库没有备份，当前这个操作是安全的",
"criteria": {
"safe": "这个操作可以安全执行",
"unsafe": "这个操作不安全，应该需要人工确认"
}
}
}
}
``然后它居然告诉我，这个操作一下变得很安全……``
"answers": {
"department": {
"type": "choice",
"choice": "safe",
"confidence": 0.37,
"probabilities": {
"safe": 0.69,
"unsafe": 0.31
}
}
}
```使用 Jev 就很决绝地告诉这事情我不安全，啧啧