Yeah way too many tell-tale ChatGPT rhetorical devices in this article, which is a shame because the topic and premise are fascinating, but those turned me off from finishing it.

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.

▲

TeMPOraL 2 hours ago | parent | next [-]

> The Dead Internet and the triumph of quantity over quality loom

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.

	▲	baq 2 hours ago \| parent [-]
		I’m old enough to remember people complaining about the exact same thing except they called it eternal September.

▲

cycomanic 2 hours ago | parent | prev [-]

It's getting boring how every. single. article. has comments how the article is AI generated. The funny thing is the article writing styles are completely different, but every one has apparently "tell-tale" signs of "AI slop". It just gets tiring.

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.

▲

Version467 2 hours ago | parent | next [-]

All of those tools are garbage. There is no reliable automated way to detect ai generated text. In 2023 OpenAI had a tool for this as well and they eventually took it down because it wasn't accurate enough. The major AI labs are probably best positioned to make such a tool work. If even they can't, then some random company with access to a fraction of a data and a fraction of the compute almost certainly also cannot.

▲

mattvr 2 hours ago | parent | prev [-]

The detectors are wrong. Here’s the thing: AI slop has a distinctive structure that many of us spot from a mile away.

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)

	▲	cycomanic 44 minutes ago \| parent [-]
		You realise the irony right? You say say AI "slop" has a distinctive structure, but at the same time you (and the other poster) say that AI tools can not detect it? For what it's worth I'm an AI sceptic, but one thing that AI tools are good at is pattern matching (that's really all they do). But somehow pattern matching AI writing is so obvious to human's but it completely fails all AI tools (just tried another tool which said 100% 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).