Recently I noticed that my responses from Sonnet 4 sounded a lot like 3.5 instead. So I did some digging around and this isn’t the first instance of this kind of thing happening apparently? Cursor better have a good response to this - breaking trust is the best way to destroy your client base.
Thats a clear hallucination. Why?
Because your screenshot shows “Thought for 4 seconds” which Claude 3.5 can NOT do.
it hallucinated because you suggested it was a different model, so It tried to please you and said its 3.5. it’s not.
1 Like
Correct, in other threads Cursor staff also explained this well.
The models know Claude 3.5 Sonnet because at time of training of the current model it was known and so it comes out as an answer. Anthropic names the models just Claude, they do not know which model they are. Thats why the name is added for clarity.
1 Like
Okay, it could be hallucinating. out of curiosity, why does it tell me Joe Biden is the president of the US then when I ask it to use the data it’s trained on to tell me who is president? When I ask claude sonnet 4 on anthropic’s website it says Donald Trump.
On Anthropics website you are communicating with their Chat tool which has its own System Instructions. One of those instructions is the last election. The fact that they added this into instructions means it might not have made it into training data.
Anthropic does not reveal which data comes from what date.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.