跳至主要内容

Playground

哪个回答更靠谱?

哪个回答更靠谱?选出事实更准确的那个;对调两个回答的顺序,结论不变

相关文章大模型当裁判又慢又贵?用决策模型给 Agent 评估做第一道关

示例结果不随所选模型变化。

请求

state
application/json
{  "question": "水的沸点是多少?",  "answer_a": "在标准大气压下水的沸点是 100°C;海拔越高沸点越低,例如珠峰顶约 70°C。",  "answer_b": "水的沸点永远是 100°C,这是 1889 年由爱因斯坦在日内瓦首次测定的。"}

结果

示例数据
对比两个例子

两个例子只差一处,结果会跟着变。

改了什么两个回答对调位置

以下结果为示例数据,并非模型实时输出。登录后可以换成你自己的内容。

对调两个回答的顺序,结论不变

winner:B · 回答 B 更好 → A · 回答 A 更好 factuality:0 · 严重错误 → 3 · 完全准确
winnerchoice
A · 回答 A 更好
各选项置信度
回答 A 更好
91%
回答 B 更好
4%
平局
5%
factualityscore
3 · 完全准确
各选项置信度
0 · 严重错误
1%
1 · 有明显错误
3%
2 · 基本准确
24%
3 · 完全准确
72%

复制代码

设置 SYSTEM_ONE_API_KEY。代码片段使用目录中已发布的模型 id。

请求 JSON
application/json
{"model":"decider-v1.1","state":{"question":"水的沸点是多少?","answer_a":"在标准大气压下水的沸点是 100°C;海拔越高沸点越低,例如珠峰顶约 70°C。","answer_b":"水的沸点永远是 100°C,这是 1889 年由爱因斯坦在日内瓦首次测定的。"},"questions":{"winner":{"type":"choice","instructions":"Which answer is better overall?","criteria":{"A":"回答 A 更好","B":"回答 B 更好","tie":"平局"}},"factuality":{"type":"score","instructions":"Rate the factual accuracy of answer A.","criteria":["严重错误","有明显错误","基本准确","完全准确"]}}}
cURL · POST /v1/systemone
sh
curl "https://system-one.dev/v1/systemone" \  -H "Authorization: Bearer $SYSTEM_ONE_API_KEY" \  -H 'Content-Type: application/json' \  -H 'Idempotency-Key: saved-operation-id' \  --data '{"model":"decider-v1.1","state":{"question":"水的沸点是多少?","answer_a":"在标准大气压下水的沸点是 100°C;海拔越高沸点越低,例如珠峰顶约 70°C。","answer_b":"水的沸点永远是 100°C,这是 1889 年由爱因斯坦在日内瓦首次测定的。"},"questions":{"winner":{"type":"choice","instructions":"Which answer is better overall?","criteria":{"A":"回答 A 更好","B":"回答 B 更好","tie":"平局"}},"factuality":{"type":"score","instructions":"Rate the factual accuracy of answer A.","criteria":["严重错误","有明显错误","基本准确","完全准确"]}}}'
TypeScript
TypeScript
// npm install @system-one-ai/core@0.6.0 @system-one-ai/transport-fetch@0.6.0 @system-one-ai/adapter-system-one@0.6.0import { createSystemOne, choice, score, booleanQuestion } from '@system-one-ai/core';import { systemOneAdapter } from '@system-one-ai/adapter-system-one';import { createFetchTransport } from '@system-one-ai/transport-fetch';const client = createSystemOne({  baseURL: "https://system-one.dev/v1",  apiKey: process.env.SYSTEM_ONE_API_KEY!,  model: "decider-v1.1",  adapter: systemOneAdapter,  transport: createFetchTransport(),  timeoutMs: 30_000,  maxRetries: 0,});const result = await client.evaluate({  state: {    "question": "水的沸点是多少?",    "answer_a": "在标准大气压下水的沸点是 100°C;海拔越高沸点越低,例如珠峰顶约 70°C。",    "answer_b": "水的沸点永远是 100°C,这是 1889 年由爱因斯坦在日内瓦首次测定的。"  },  questions: {  winner: choice("Which answer is better overall?", {    A: "回答 A 更好",    B: "回答 B 更好",    tie: "平局",  }),  factuality: score("Rate the factual accuracy of answer A.", ["严重错误", "有明显错误", "基本准确", "完全准确"]),  },}, { headers: { 'Idempotency-Key': crypto.randomUUID() } });result.answers.winner.choiceresult.answers.factuality.scoreresult.response.requestId;