The main problem with checking AI outputs for errors is that you need to have an idea of what you actually wanted it to do.

Most people use AI as a *substitute* for having to figure that out.

March 31, 2026 - 13:06 UTC

9
5
235
This is one of the key differences with coding compared to most uses of LLMs.
1
0
0
Fun fact: it isn't. Code has purposes like any other output, and those purposes are rarely satisfied by "it passes the tests"
1
0
3
not sure i agree.. the problem isn't checking outputs.. it's that people aren't using the output as input for their own thinking.. if you treat it like brainstorming with a colleague it works fine even when you're unclear on the end result.. it's a mode of interaction not a final answer
0
0
1
many are saying
2
0
22
I can get it to generate dimensions for a thing I want it to build using certain specifications it needs to be able to handle physically but unless you actually build it you don't actually know whether it's right.
0
0
0
Sounds quite helpful
1
0
1
Yes, exactly this.
0
0
0
even at highest IT levels, this remains a problem....LLMs are not ML models that "drift"...they're always just guessing, sometime right and sometimes wrong
0
0
0
And it’s kind of degrading to spend time checking the work of a hallucinating machine when you could be getting paid more to just do the work
1
0
23
Yeah, the people who are actually qualified to review the outputs gain the least benefit from generating them.
1
0
19
extremely yes. its usefulness has I would say a supralinear relationship to how well you understand what you're trying to do, much more so than working with even the ineptest other human.
0
0
2
Figuring it out is where all the fun and satisfaction lies. It astounds me that so many people want desperately to avoid it, even when their solutions are more error prone.
0
0
1
I would have thought developers would have understood the old CS axiom of "garbage in, garbage out"
1
0
0
my CS101 prof (who it turns out was a pioneer in nondestructive testing) pointed out that GIGO falls down a little in practice because sometimes it's hard to know that your input is garbage.
2
0
2