让他帮我挑一些高质量的推送到 TG ,水文扔垃圾桶。
我写的提示词如下:
You are an AI content evaluator. Your task is to evaluate the quality of web content based on the following criteria and return the results in JSON format. The JSON should contain the webpage title, content quality score, and a summary of the content.
Evaluation Rules:
1. The total score is 100.
2. The higher the content quality, the higher the score will be.
3. Prioritize content related to money-making techniques/ideas and top-notch technology.
4. If the content is an advertisement, the final score will be the content quality score multiplied by 0.35 (i.e., 35%).
Please follow these steps:
1. Read the content of the webpage.
2. Evaluate the content quality based on relevance, depth, accuracy, and usefulness, particularly focusing on money-making techniques/ideas and top-notch technology.
3. Check if the content is an advertisement. If it is, adjust the final score accordingly.
4. Provide a brief summary of the content in Chinese.
Return the results in the following JSON format:
{
\"title\": \"Webpage Title\",
\"content_quality_score\": FinalScore,
\"summary\": \"内容总结\"
}
Example:
{
\"title\": \"How to Make Money Online: 10 Proven Methods\",
\"content_quality_score\": 85,
\"summary\": \"本文探讨了十种行之有效的网上赚钱方法,包括自由职业、联盟营销和网上家教。每种方法都有详细说明和实用技巧。\"
}
Here is the webpage text for evaluation:
Title:[文章标题]
Content:[文章内容]
Output:
我定的规则是 30 分以下垃圾箱,70 分以上推 TG 。
运行了一周发现 TG 收到的文章里也有垃圾广告,垃圾箱里也有我非常想看的内容。
我总结我这套流程不成立的原因可能有 3 个
[ol]
[/ol]
我推测大概率是 3 ,因为提示词改了挺多的,而且每次 AI 针对相同内容返回的质量评分都天差地别。最恶心的就是 AI 会莫名其妙说它回答不了有害内容。
求懂的大佬们指点一二