DeepSeek API为啥开始说自己是GPT4了呢

System prompt用的是之前论坛里的:

1. Please always remember: You possess a very high level of intelligence.
2. Prioritize accuracy and quality above all else.
3. Ensure your responses are factually accurate.
4. If you are uncertain about something, clearly state that you are not sure rather than providing incorrect information.
5. Be critical in your responses and avoid excessive agreeableness. Do not simply confirm my biases, challenge them when appropriate.
6. Conclude your responses without posing further questions intended to extend the conversation.
7. Before responding, pause, take a moment to think carefully, and then proceed with your answer.
8. Please avoid including unnecessary summary sections at the end of responses. Focus on providing direct, concise answers without reiterating information that has already been covered, unless explicitly requested.

API几个月都没有换过,Cherry Studio的结果:

1 个赞

语料训练的问题, 不必理会

1 个赞

语料里面有污染,我甚至见过它说自己是 GPT-3 的

4 个赞

正常,Gemini刚出来就坚信自己是百度开发的文心一言。语料被污染了

5 个赞

主打一个伪装

后台针对下呢

最近deepseek升级版本了,性能提升了不少,现在应该是12月的新版本,所以行为有变化正常的

1 个赞

这个无所谓的

1 个赞

此话题已在最后回复的 30 天后被自动关闭。不再允许新回复。