Author here. I ran into this while comparing Pendulum with `whenever`, the datetime library I maintain. I sent a patch that removes the hack, but the post is about why the underlying design can't be patched the same way. Happy to answer questions.

▲

LoganDark 3 hours ago | parent [-]

Did you use LLM to write parts of the article?

> The trouble with guessing who’s calling is that you can’t anticipate every caller. Pendulum’s hard-coded check is fragile: it covers only one name, and relies on an exact call stack that Pendulum doesn’t control.

Edit: it looks like the later parts of the article show a lot of LLM tells, but not the beginning. I've been seeing this pop up more and more lately...

▲

seki285 2 hours ago | parent | next [-]

This is impressive, because I had no idea and now I feel angry at having my time wasted! If this type of stuff increases in frequency and I can't easily tell if an article is AI or not, I will stop reading them.

▲

awildfivreld 2 hours ago | parent | prev | next [-]

Up until "Each guess may look reasonable on its own. But guessing itself has four problems." I read the occasional tell, but not much.

By the end of that paragraph thereafter I started skimming as the rest turned blatantly LLM.

I should have caught the emdash in the first code block in "Who else is calling" though...

▲

throwaway89201 2 hours ago | parent | prev [-]

> but not the beginning

You didn't stop reading after "The hack itself is on its way out, but the design behind it still matters for your code"? I did.

▲

LoganDark an hour ago | parent [-]

I meant the beginning didn't show a lot! I don't have exactly zero tolerance because sometimes there's enough signal in the slop, my point was mainly that the signal seems to sharply drop off after a certain point in this article. (I've complimented authors before for managing to keep an article relatively information-dense despite having a lot of clear LLM tells. That shows care, and so I don't object to it.)

I have a suspicion that in these cases, either people are purposefully copyediting only the very beginning, or the very beginning has most of the prompt and the rest is pure extension.

	▲	ok_dad 27 minutes ago \| parent [-]
		Or they write a short article or a rough draft, and get the clanker to zhuzh it up. I've experimented with clankers doing writing for me, but they freaking suck so I refuse to do that any longer, and what I noticed is they tended to just change a lot of what I wrote but then extended the writing with no-meaning bullshit and removed all of the soul from my writing at the same time. IMO: writing ~= thinking and if you let a clanker write for you then you'll forget how to think properly, so no one should be doing it.