Stack Overflow data reveals the hidden productivity tax of ‘almost right’ AI code
Stack Overflow data reveals the hidden productivity tax of ‘almost right’ AI code
venturebeat.com
Vercel Security Checkpoint
9 comments
-
-
‘Almost right’. I prefer the term: confidently incorrect.
-
It's ironic that this comes from Stack Overflow where I've seen on more than one occasion the wrong answer selected as the solution whilst the right answer was ridiculed and voted down.
This was happening long before Assumed Intelligence was pretending to be the nail to every hammer.
-
do you have an example of this?
-
-
Invest in debugging and code review capabilities: With 45% of developers reporting increased debugging time for AI code, organizations need stronger code review processes. They need debugging tools specifically designed for AI-generated solutions.
Or, maybe, don't use tools that generate garbage code.
-
"2024 the report found that developers were not worried that AI would still their jobs." Still? I think AI might take the articles author job first.
9 comments
I prefer the term slot machine coding because you keep thinking you almost have a jackpot but there's one lemon in there so you get that rush to keep pulling the lever and expecting different results.
That's a great way of putting it and definitely something I was guilty of doing when this stuff first emerged and I was experimenting with it.
Nowadays, I only use our internal LLM to generate boilerplate or simple scripts that wouldn't take me more than 5-10 minutes to write myself to save some tedium. I think that's what most actual devs/admins do with it nowadays, if they interface with the tech at all.
This really puts into perspective why everyone seems to be so addicted to asking the AI bullshitter questions. Once again, the problem is gambling