Once again, a promising article is completely ruined by blatant ai-isms. I could only make to the end of the pointer section before I couldn't take it anymore.
There is a real crisis of AI slop getting posted to this forum. I don't even bother reading posted articles related to AI anymore, but now it's seemingly extending to everything.
"This means reading light data requires zero locks. No mutex, no spinlock, nothing." threw up red flags, and by the time I got to "But here’s the insight" I couldn't go any further.
I’ve been trying to put my finger on what gives it away. It’s that there are boolean trees underneath each text decision it makes. While humans are obviously capable of that, our conclusions and framing are more continuous. This why you for example see LLMs constantly defining things by what they’re not.
The main issue is sota LLMs can only reason one way - forwards, and can't go back and revise a prior statement. That would remove a whole lot of "it's not this is that" and "the big takeaway here is" and so on. Those kinds of ideas are typically at the beginning of a human writer's output structure. An LLM can't go back and edit the first paragraph, because it has to reason (for whatever that means for LLM) it's way through it to get to the big idea of the paragraph/structure. I haven't played with diffusion text models enough to know if that's a remedy for that kind of output.
When LLMs are good enough to not be detectable, what happens then? They aren't that far away atm, so it's only a matter of time until _everyone_ is assumed to be an LLM.
LLMs are trained to be precise (and more specifically: semantically precise), especially in the fine-tuning phase. An LLM just trained on the corpus of full human production would surely sound more "human," but it would also probably be pretty useless. So that's why idioms like "it's not X, it's Y" are a dead giveaway; but really, any structure that tries to "guide" our salience is a dead giveaway. Here's a random paragraph from Knuth's Literate Programming†[1]:
> For example, a system program is often designed to gather statistics about its own operation, but such statistics-gathering is pointless unless someone is actually going to use the results. In order to make the instrumentation code optional, I include the word ‘stat’ just before any special code for statistics, and ‘tats’ just after such code; and I tell WEAVE to regard stat and tats as if they were begin and end. But stat and tats are actually simple macros.
I encourage you to read that paragraph a few times. Even if you have no idea what the context is, you get that there's a point, that there's something else to dig into, that the author might be being a bit cheeky. In other words, you can feel Knuth behind the ink. Philosophers would call this intentionality[2]. LLMs produce the polar opposite of garden path sentences[3] (and, imo, that's why they're so easy to spot).
† I specifically picked something technical to illustrate that even in domains where semantic precision is of utmost importance, human expression is still just that: human.
Hey, the author of the blog here. English is not my native language. I did try to write the first section, but for the later section, I wrote a rough draft, but then I had that intrusive thought that people would not like it because of my poor English and too many typos, so I just told LLM to make it better, and that's how it ended up like this. While my intentions were right, my method was wrong. I will improve and fix my writing. I have taken down the blog for now. I don't want to waste other people's time. While it has good content, the presentation is wrong!
i would genuinely rather read the rough draft before it got turned into this slop. it would be messier, maybe, but it’d have actual human insight and direction.
19 comments
There is a real crisis of AI slop getting posted to this forum. I don't even bother reading posted articles related to AI anymore, but now it's seemingly extending to everything.
When LLMs are good enough to not be detectable, what happens then? They aren't that far away atm, so it's only a matter of time until _everyone_ is assumed to be an LLM.
> For example, a system program is often designed to gather statistics about its own operation, but such statistics-gathering is pointless unless someone is actually going to use the results. In order to make the instrumentation code optional, I include the word ‘stat’ just before any special code for statistics, and ‘tats’ just after such code; and I tell WEAVE to regard stat and tats as if they were begin and end. But stat and tats are actually simple macros.
I encourage you to read that paragraph a few times. Even if you have no idea what the context is, you get that there's a point, that there's something else to dig into, that the author might be being a bit cheeky. In other words, you can feel Knuth behind the ink. Philosophers would call this intentionality[2]. LLMs produce the polar opposite of garden path sentences[3] (and, imo, that's why they're so easy to spot).
† I specifically picked something technical to illustrate that even in domains where semantic precision is of utmost importance, human expression is still just that: human.
[1] https://www.cs.tufts.edu/~nr/cs257/archive/literate-programm...
[2] https://plato.stanford.edu/archives/fall2014/entries/intenti...
[3] https://www.sjsu.edu/writingcenter/docs/handouts/Garden%20Pa...
> Its shit too, but our kind of shit.
Unfathomably based.