Edit: old post is someone is curious: https://dev.to/jamilxt/electron-vs-gpui-what-discords-290-mb-rust-rewrite-can-and-cannot-prove-34b6
Replaced with the one from The Verge
Edit: old post is someone is curious: https://dev.to/jamilxt/electron-vs-gpui-what-discords-290-mb-rust-rewrite-can-and-cannot-prove-34b6
Replaced with the one from The Verge
LLM writing, just infuriating to read.
Sorry, I missed it.
I spend 10 minutes trying to find an article about that news, the only other option being the post by the Discord CTO on Twitter of all places.
I’ll see if there’s a genuine blog post about it now.
Edit: updated to Discord is ditching Electron and its new app uses less RAM
Another example of how a human could have delivered the same information in 25% the time.
The LLMs have been trained on articles written by humans, who, in turn, have been trained to draw out their conclusion into several pages in order to fit advertisements between the paragraphs and pages. It was infuriating before LLMs, and it’s equally infuriating now.
One reason I like Kagi is that it downranks ad-heavy pages. And of course I like uBO. But there are still so many web users without ad blockers that I makes sense for sites to spew slop, just like you’re saying. Blows my mind that anyone would use the web without a blocker.
Before LLMs were a thing, I would just out of sheer muscle memory skip the first few paragraphs of an article to get to the point of what the headline was about, because every fucking article starts with the big bang and works its way to the content of the headline.
I’m getting worse at spotting these. What gave it away?
LLMs use the wrong language cues. They spend a lot of words emphasizing an utterly mundane point, and it’s difficult to read because it’s constantly winding you up to reveal some arcane knowledge just to drop knowledge bombs like, make sure you use good names for variables and always handle your exceptions. I’ve seen them spend two paragraphs when a short sentence would do.
I think they are improving on this too, some latest models give pretty terse answer for example
Just generally it waffled on, and didnt get to a point.
Thats the “its not X, its Y” pattern.
Plus all the other constant flip/flop contrasting throughout the text. It’s like a comparative decision matrix converted to prose, and exhausting to read.
Not to mention the ai writing assistant pop up or whatever’s this is upon viewing the page:
Fortunately I disallow javascript execution by default, so I often don’t see this crap. Of course, about half the time I don’t see images or the article content either, since delivering text or images is apparently completely impossible without running arbitrary code.
I seem to vaguely recall we used to have some sort of hyper-text markup language that could do all that, but I guess I must’ve imagined it.