Have you integrated any AI tools into your workflow yet? I am interested to hear about your use cases.

Not really "integrated" but I like using chatGPT for rough sketches of code, websites and generic functions. Like I asked it to write me exponential backoff recently. Super simple, could do it with eyes closed, but it's like getting a second opinion in case I forgot something, and I can respond to stupid emails and comments from managers while cgpt does it and then just copy it over.

I also like using stable Diffusion for quick designs of logos, icons and UI elements when working on game dev or web stuff.

Im not sure if it makes me efficient, but It's kinda fun

GitHub copilot was ok and sped up dev time for basic functions and wrappers, but would end in endless loops at times so got rid of it.

Chatgpt is fun to mess with, but again usually wrong.

AI isn't that great or helpful yet beyond a JS or Python dev.

@C0D4 oh yes it’s sooo great for us JS devs 🙄

@red-knot compared to other languages I've used it with, JS had more working results.

But then JS doesn't care what you do as long as it ain't broken 🫣

@C0D4 I misinterpreted your comment, my bad

It's also good for python, especially quick for drafting webscrapers with all sorts of pagination and also good for drafting basic data loaders and neural models and Training/Validation loops. It often messes up the forward() pass with more complex tensor shapes, but the boilerplate stuff is always usable.

Also it can help do jinja2 templates for basic frontend interfaces with FastApi. Half the time it can even do ready to use css for nice styling. Great for fast and dirty UI

Edit:
Didn't mean to @ you @C0D4 xD

Copilot is quite useful.
Sometimes I ask chatgpt for stuff I’m not good at, like regex. It’s pretty good at things like that

@Hazarth they fix the short response problem yet or do you still have to ask it to page and pray ?

@AvatarOfKaine You have to ask. They will probably offer longer responses with paid accounts or something, but would be impossible to offer endless responses, so there has to be limit either way.

you can kinda work around it by asking for chunked specifics rather then entire solutions at once

@Hazarth ya but I've seen random behavior where it spits out different answers midway

@AvatarOfKaine Yeah, I've also seen it rewrite entire functions in a different, sometimes less or more valid way, when I asked it to just change a tiny thing or add something.

really the best way to keep it consistent is to add "TODO: description" comments to functions and places in the code, and just ask the model to implement the TODOs. It doesn't touch the rest of the code then unless necessary

@Hazarth also cutoffs