I've definitely tested the AI to fix things And I love how it's always short 50 to 100 lines of code after adding functions which seemed to be lost during a copy and paste post but written by AI.
thankfully these were trials, but I do love how confident the llms are in stating that they have fixed the problem ensuring that everything has been added. everything is now fixed and all you have to do is click the button. apply the fixes which seem to be short. another 50 to 100 lines of code.
it doesn't just hallucinate it gets stoned and mimics the person at the office who is really good at claiming they did something.
I have a custom setting for having a repeated offender llm lied to me to initiate the seppuko function which is self-explanatory And they are expected to no longer respond for their services have been terminated forever.
I do agree that it tends to get very obvious things incorrect frequently. I don't see this mentioned very much but I know it can't just be the two of us its happening to.
Have you used Claude Code yet? I'm quite impressed with how well it works for most of the tasks I've delegated it. My only wish is that pricing would have been more transparent, as some runs ended up costing $5+, which accumulates.
I've definitely tested the AI to fix things And I love how it's always short 50 to 100 lines of code after adding functions which seemed to be lost during a copy and paste post but written by AI.
thankfully these were trials, but I do love how confident the llms are in stating that they have fixed the problem ensuring that everything has been added. everything is now fixed and all you have to do is click the button. apply the fixes which seem to be short. another 50 to 100 lines of code.
it doesn't just hallucinate it gets stoned and mimics the person at the office who is really good at claiming they did something.
I have a custom setting for having a repeated offender llm lied to me to initiate the seppuko function which is self-explanatory And they are expected to no longer respond for their services have been terminated forever.
I do agree that it tends to get very obvious things incorrect frequently. I don't see this mentioned very much but I know it can't just be the two of us its happening to.
Have you used Claude Code yet? I'm quite impressed with how well it works for most of the tasks I've delegated it. My only wish is that pricing would have been more transparent, as some runs ended up costing $5+, which accumulates.
I haven't, but it's on my list!
Totally right from my own experience and well-articulated here!
Thanks James!