I’d still be highly sceptical about pull requests with code created by llms. Personally what I noticed is that the author of such pr doesn’t even read the code, and i have to go through all the slop
Ya I’m finding myself being the bad code generator at work as I’m scattered across so many things at the moment due to attrition and AI can do a lot of the boilerplate work, but it’s such a time and energy sink to fully review what it generates and I’ve found basic things I missed that others catch and shows the sloppiness. I usually take pride in my code, but I have no attachment to what’s generated and that’s exposing issues with trying to scale out using this
Same. There’s reduction in workforce, pressure to move faster, and no good way to do that without sloppiness. I have never been this down on the industry before; it was never great, but now it’s terrible.
Couple of years back I got a PR at work that used a block of code that read a CSV, used some stream method to covert it to binary to then feed it to pandas to make a dataframe. I don’t remember the exact steps it did, but was just crazy when pd.read_csv existed.
On a hunch I pasted the code in google and found an exact match on overflow for a very weird use case on very early pandas.
I’m lucky and if people send obvious shit at work I can just cc their manager, but I fell for the volunteers at large FOSS projects, or even paid employees.
I’d still be highly sceptical about pull requests with code created by llms. Personally what I noticed is that the author of such pr doesn’t even read the code, and i have to go through all the slop
I suspect the answer will be that such large requested as you frequently see with LLM codegen will just be rejected.
Already I see changes broken up and suggested bit by bit, so I presume the same best practice applies.
Ya I’m finding myself being the bad code generator at work as I’m scattered across so many things at the moment due to attrition and AI can do a lot of the boilerplate work, but it’s such a time and energy sink to fully review what it generates and I’ve found basic things I missed that others catch and shows the sloppiness. I usually take pride in my code, but I have no attachment to what’s generated and that’s exposing issues with trying to scale out using this
Same. There’s reduction in workforce, pressure to move faster, and no good way to do that without sloppiness. I have never been this down on the industry before; it was never great, but now it’s terrible.
Some thought I had the other day :LLM is supposed to make us more productive, say by 20%. Have you won a 20% pay rise since you adopted it? I haven’t
Did we all forget about stackoverflow?
Peopleblindly copy/pasted from there all the time.
Couple of years back I got a PR at work that used a block of code that read a CSV, used some stream method to covert it to binary to then feed it to pandas to make a dataframe. I don’t remember the exact steps it did, but was just crazy when pd.read_csv existed.
On a hunch I pasted the code in google and found an exact match on overflow for a very weird use case on very early pandas.
I’m lucky and if people send obvious shit at work I can just cc their manager, but I fell for the volunteers at large FOSS projects, or even paid employees.