For some reason, for the first time for me, the AI has hallucinated and responded with weird results. It seems to think that I responded, hallucinating my responses, and replied to those hallucinated responses. It kept replying to the imagined messages until I hit ‘Stop’.
I was using claude-3.7-sonnet, using both thinking and Max mode.
Hm, this is actually happening fairly frequently. It only started happening after updating to 0.48.X. It may be specific to the Max, or thinking, version of Claude 3.7 Sonnet – or the combination of both – but I’m not sure.
It seems to hallucinate not just my responses to its messages but also seems to think it doesn’t have access to my files.
Versions
- Cursor: 0.48.6
- VSCode: 1.96.2