I could use AI to edit my podcasts. As well as the ethics, the reason I don't is because it wouldn't save me any time. I'd still have to go through and check everything, and fix the small mistakes it made.
People who say AI saves them time at work (coding for example) must not be thoroughly checking the output and fixing the small mistakes. Or maybe I'm missing something here. 🤷‍♂️
@joeress You’re not missing anything.
I’ve used Claude professionally since about Oct last. I’ve manually written very little code since.
But I spend an inordinate amount of my time reviewing the output and issuing more prompts to fix things, and then reviewing again.
I was impressed at the start, but now I view it as a random bug generator. And it’s making me lazy as a coder.
So where’s the upside? I’m not sure there is one :-)
@t3rcermillenium @joeress does it actually save you time overall? Why do you bother with it if all it does is give you more tedious fixing when you could have put that effort into doing it right to start with? Seems like a good way to make your job far more annoying for no real gain
(I’m curious - I haven’t used it and am not exactly convinced…)
@mich181189 @joeress I wrote a desktop app in a GUI framework I had zero experience with (and tbh, I’ve never actually written a full GUI app - I’m mostly a backend dev).
Doing it without Claude would have been very painful.
However, if I developed it myself, I’d now know a lot more about that UI framework. As it stands, we have a new app in production, and a very experienced dev who has learned nothing new :-)
@mich181189 @joeress Overall, I seem to trade one activity (manually writing code) for another (reviewing and finding bugs).
But it’s worse. Eg: user pings me about a bug, I walk over and see bug. Normally, by the time I’m back to my desk, I’ve a good idea what’s happening. Now? No idea. I didn’t write the code so have only vague ideas. So now I use Claude to debug.
Cognitive debt is the new tech debt.
Why use AI? It’s being push by mgmt :-(
@t3rcermillenium @joeress I have a very similar experience. As is typical, I recently used it to write a simple use case of C++ 17s recursive_directory_iterator. After a little testing I discovered it crashes if it runs across a directory it doesn't have permission to open, a very normal use case. That's great behavior for a backup utility. After an updated prompt it crashed on undefined behavior using an invalidated iterator...
Fortunately I only use it to make fun of it, not professionally. I always think about how many people use output without the level of checking we are doing and that scares the crap out of me.
@kookjr @joeress This rings so true.
The code it generates looks good, has most of the edge cases covered, but not all edge cases.
It’s like “where’s Wally” trying to find the missing edge cases in the generated code.
At least when I’m coding, I’m thinking about the edge cases. So I still have to think about the edge cases, but now I also have to find the ones missing in the code.
Crazy!
Joe Ressington
Ewon
T3rc
Michael
Mat Cucuzella