Post #4377498

2026-08-04 14:04 UTC

No disrespect to Schneier, but "job vs gym" is an argument about automation vs exploration, not an argument about AI. Sure, automate what can be automated. Get yourself forklift certified. But the real question when it comes to "should I do this myself or should I let an unreliable text generation tool do it" is not "human or the LLM". If poorly or unoriginally or unreliably is an option the question is "does this need to be done at all." https://www.schneier.com/blog/archives/2026/07/should-you-use-ai-for-a-task-heres-a-simple-way-to-decide.html

Replies (4)

@wendynather@infosec.exchange 2026-08-04 14:30

@mhoye@cosocial.ca What’s wrong with a good old-fashioned shell script?

Open ##4377496
@mhoye@cosocial.ca 2026-08-04 14:29

I periodically think about some of John Carmack's observations about functional programming as technique and style, an intertwined set of ideas for which we need a better term, and mull about the fact that I can see (but am not smart enough to implement) an equivalent set of arguments about stochastics. e.g. "If something is close to deterministic or well parameterized, make it fully deterministic." (There's no AI equivalent of "minimize control flow complexity", obvs) https://cbarrete.com/carmack.html

Open ##4378499
@froztbyte@mastodon.social 2026-08-04 19:17

@mhoye@cosocial.ca counterpoint: schneier really doesn’t deserve as much of the respect (and leeway) he gets if you go looking, the dude’s got a multi year track record of being out of touch with the now and then some-weeks-later posting some almost-correct silliness that still falls on its face there are better voices to follow

Open ##4387073
@tbortels@infosec.exchange 2026-08-04 15:38

@mhoye@cosocial.ca There's a false dichotomy at work here: "do it myself" or "let the AI do it unreliably" are not the only options. Rather than spend 4 hours cutting and pasting boilerplate do do something I've done 8 times before at another employer, I can have the AI take 10 minutes to put together that same boilerplate - likely from the same sources I use - and then spend an hour turning the unreliable result into reliable, tested code. You don't do "either-or", you do "and". Then spend the 3 hours you just saved doing something more creatively interesting. The AI isn't a replacement - it's a tool. It accelerates. And yes the output is often crude or just wrong - correctness (and planning and fit-for-purpose and budget and about 18 other things) is your job. A chainsaw is dangerous, and can cut down the wrong trees. It's your job to mitigate that danger and choose the right trees. The real issues with AI are not as simple as "it's wrong, a lot". It's issues of training and intellectual rights and what it's doing to the economy and displacement of workers and *none* of those are the AI's fault - they're the fault of human beings creating and selling them. As is crappy output making it past the human who should have fixed it.

Open ##4389603