AI slop hits 700+ upvotes on Hacker News. The Dead Internet and the triumph of quantity over quality loom. A sign of things to come.
The key takeaways - that solar is super cheap, that technology has unlocked offering hire purchase to incredibly poor remote communities, and that westerners looking to buy carbon credits are basically subsidising that - are pretty interesting. There was a lot of fluff around those points though.
always_has_been.jpg
The Internet has drowned to death in garbage back when they coined the term "content marketing". That was long before transformer models were a thing.
People have this weird impression that LLMs created a deluge of slop and reduced overall quality of most text on-line. I disagree - the quality arguably improved, since SOTA LLMs write better than most people. The only thing that dropped is content marketing salaries. You were already reading slop, LLMs just let the publisher skip the human content spouter middleman.
Yeah, I'm old.
I dunno if HN is at that point yet, but it's certainly creeping closer compared to where it was 5-10 years ago. Reddit passed the point of no return within the last few years.
For what it's worth I pasted the first couple of paragraphs into several AI detectors and 4/5 said it's clean, while one said mixed (partly AI generated partly human). So either all these AI generation tools are crap, or the text is not so "obviously" AI generated. Not saying either way, but it seems to at least not be so obvious.
[0] https://mattsayar.com/can-ai-tell-if-im-writing-ai-slop-a-ma...
[1] https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
The kicker? This setup-punchline format sets off a red alert for astute readers’ AI detectors.
This isn’t just AI slop, it’s an industrial AI sludge factory.
(note: this was ironically written by a human)
It doesn't match up. Moreover it's getting tiring, because every single article has these comments on them, and I've seen enough examples where authors showed up in discussions or texts were from before LLMs were widely available, but posters were still adamant that the text was AI generated.
I highly doubt that people here would reliably pick out (success rate > 60%, i.e. you get 60% of guesses correctly if text was generated by a human or LLM) LLM generated text that completely fools 90% of AI detectors.
Regarding the setup-punchline format, guess what, those were popular way before LLMs (not surprising LLMs must have learned them from somewhere).
