In my experience, I’ve encountered the same issue with claude-3-5-sonnet-20241022 as well. It’s not that all Chinese characters are corrupted—only a small portion of them. However, it’s exactly those few that often lead to inexplicable bugs. It would be great if this could be addressed as well.
The Claude series models do occasionally display garbled Chinese characters, but it does not affect the overall use. The Gemini series has garbled characters in every answer and is completely unusable.
Thank you! I have now given up on using Gemini provided by Cursor, and instead directly input Gemini’s official API Key into Cursor. There will be no garbled Chinese characters when filling in the official API Key.
It’s not just the Gemini model, I’ve often encountered the problem of occasional Chinese characters turning into the unknown symbol ‘�’, as in the example below, which ai keeps trying to fix, but it’s not working.
The problem is NOT the AI model, is the cursor built in encoder! After the generation, cursor will compare and match and replace the corresponding text in your file, during that process, the garbled text “��” will randomly occurs across all of your files!!
This is devasting for indie devlopers since it is like minefield, can happen very randomly, you can prevent this no matter how good your prompting skills are!!
There is no problem with Claude. The problem is the ‘apply’ of Cursor, which has a small probability of generating Chinese garbled characters. However, the problem with Gemini is that 100% of the replies on the Chat page will generate Chinese garbled characters, and Gemini is unusable.