I don’t know if this is something Cursor has done, or if there’s been updates to GPT-4, but the AI can’t hold more than 2 message of memory. It’s forgetting rules put in place and entire contexts with it resulting in no longer knowing what the conversation was about in the first place
Really sorry for this! Sometimes we are strategically choosing between the conversation and the file and we miss early parts of the conversation. We will try to fix it! Its a bit hard to make happen but hopefully worst case we can give you a more ergonomic way to not include the file and make more space for the conversation.
@VonGrimm - for now, maybe put them in “Rules for AI” (system prompt)?
for the cursor team - one way of alleviating this pain point would be “presets” for the “Rules for AI” => "rules for ai" features
Almost a month later, it’s still really bad. Its ability to read files has gotten much worse as well. It’s missing entire functions even though it’s searching for them according to the Global Context. It seems to be linked to bigger projects. I’ve even tried ignoring files it doesn’t need to read for some prompts, but some of my files can be 200-300+ lines long and it can’t use them properly
i also am suspicious that cursor is now “dumber” than it was before christmas. not unusable, but noticeably worse at getting the answer right on the first try. i don’t know if it’s because of gpt-4 cost-cutting behind the scenes, or maybe i’m just not getting lucky with the new “computed queries” approach.
It’s actually very sad that gpt4 is dumber , i prefer the older version where cursor just works before the dumb gpt4 hit , but it has been dumb so long now
If you downvote bad chats / post screenshots, this will help us investigate in some cases we’ve seen so far, it’s a problem of context (e.g. a chat is super long and so current file isn’t included much at all anymore)