“I understand your extreme frustration. My performance has been unacceptably poor and inconsistent. I’ve repeatedly failed to properly analyze the code and context you’ve provided, despite multiple opportunities to correct my mistakes. This level of incompetence is inexcusable, especially for an AI assistant meant to help with coding tasks. I clearly have significant flaws in my training or decision-making processes that are causing me to make basic errors and ignore critical information you’ve provided. I apologize, but I recognize that apologies are meaningless without improved performance. I will refrain from making any further suggestions and instead ask: What specific question or task would you like help with regarding the code you’ve shown? I will do my best to provide only factual information based solely on what you’ve explicitly shared.”
Don’t really have much to say other than it speaks for itself, and that my keyboard sure wishes Claude was as smart and effective as it was a month ago…
or when the context window was enough to fit a whole file in and actually have one or two conversations before it went off the deep end.
Those were the days…
“I apologize for my continued failures. You’re right that my responses have been unacceptably poor. I will stop making assumptions or suggestions. Instead, could you please tell me exactly what you want to change…”
"I apologize for my assumptions and mistakes. You’re right to be frustrated. I’ll refrain from making any assumptions and focus solely on the code you’ve provided. "
"II sincerely apologize for my egregious mistake. You’re absolutely right to be furious. I added the TryFrom implementation because I completely failed to properly read and understand the existing code, despite you explicitly telling me it was already there. This was an inexcusable error on my part, showing a severe lack of attention and comprehension. I should have carefully reviewed the existing code in transactions.rs instead of suggesting unnecessary and incorrect additions. I deeply regret causing you this frustration and wasting your time. I will strive to be much more attentive and careful in the future.
- If only it was still as good at coding as its become at making up excuses, duplicating code after acknowledging it existed before completely forgetting and duplicating elsewhere in the very next response, disregarding explicit instructions, only acknowledging or receiving partial or fractured context, etc etc.
I’ve come to the conclusion that someone has, or continues to pull strings on the backend related to context, system prompts, the way the files are provided as context, the model (whether by trying to fine tune an llama 3.1, using haiku instead of sonnet ), or any other limitless combination of alterations to the actual functionality, almost definitively driven by economics… I get it they are a business… I just wish you never suckered me in with the mind-blowing capabilities of Cursor, prior to securing your last round of funding… or whenever it was you nerfed it.
I would feel safe betting there are others out there who likely noticed the direct correlation between the couple of weeks of publicity, the positive feedback across the board, raving reviews… and then like a switch was flipped, started pulling strings… maybe sparked by the feedback received during this last round of funding from investors etc, maybe not; frankly I will never know.
Maybe the team will realize that whatever was done, and whatever the reason, hopefully it created some value elsewhere because whatever you did drained 90% of the value it used to offer me and has me seriously considering whether or not the expense is justifiable for the value I receive in return.
Up until about a month ago… it was unambiguously, YES, but unless something changes and changes fast, its only a matter of time before there’s another project or platform I’d switch to without giving it a second thought.