Conversation
Notices
-
Embed this notice
I'm trying out JetbrainsAI. If it proves to be good I'm probably gonna go ahead and just cancel copilot because it's been jeeted to shit for months now
- Fish of Rage, Johnny Peligro and pwm like this.
-
Embed this notice
@RustyCrab wanna know
-
Embed this notice
@RustyCrab honestly just the regular auto complete is so good these days I'm not even tempted to try that
-
Embed this notice
@RustyCrab maybe one day but
-
Embed this notice
@pwm it is good but it can't generate multilinear boilerplate (that's not really what it's for)
I'm currently writing coroutines in C++ and that's a good example of where you need a pretty large amount of obnoxious boilerplate
-
Embed this notice
@RustyCrab yeah I'm, not doing that
Last project was a multi threaded async task lib in python and it almost always could guess "what comes next"
-
Embed this notice
@RustyCrab hehe
Poop
-
Embed this notice
@pwm write a LOOP
-
Embed this notice
@pwm so far this is beating the fucking shit out of copilot. I just wrote a comment that said "write a poop that waits on this condition variable for one microsecond and then yields the thread" and it nailed it. Even put in a nice timer for performance profiling.
-
Embed this notice
@RustyCrab I just wish they would get off their asses and finish fleet already
There is no spell check
In a text editor
That edits
Text
Couple other annoyances but that's the dumbest one /shrug
I've done their survey like 3 times, still no spellcheck
-
Embed this notice
@RustyCrab do they let you run their model locally or do you bring your own and point the ide at it?
-
Embed this notice
@pwm real talk though I'm already impressed by this. I might look into setting up local completion because my machine is powerful enough for that I think.
-
Embed this notice
@pernia @RustyCrab some of us write code for a living (not me)
-
Embed this notice
@RustyCrab more like jeetbrains, use acme(1)
-
Embed this notice
@RustyCrab but not their model? Wouldn't that not be the same?
-
Embed this notice
@pwm it says you're supposed to use a locally hosted llmstudio or ollama server and put in the ip/port number
-
Embed this notice
@pernia @RustyCrab use ed, as of the 1970s it's the standard (and therefore the best) text editor for unix
-
Embed this notice
@RustyCrab :niggathonk: I assumed they would have at least taken an open one and trained it further with some secret sauce but perhaps you're right. I haven't read what they say they are doing on the backend
-
Embed this notice
@pwm I doubt they even have their own model. It's probably using Claude or something. Models take billions of dollars to train.
-
Embed this notice
@pwm me neither.
This is interesting though. It's a lot less eager to suggest things than copilot is, but the suggestions themselves are MUCH higher quality... obviously so.
That used to be an acceptable tradeoff but copilot has degraded hard in the past few months to the point that it seems no longer usable for C++
-
Embed this notice
@pwm this is honestly a pretty incredible difference.
Once again Microsoft starts on top and then degrades to the bottom of the barrel by virtue of simply not caring.
-
Embed this notice
@nyanide @RustyCrab ed pmo