Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Drowning in AI Generated Garbage: the silent war we are fighting

ploum.net

Post reply on HN
Print view View original

151–160 of 452 posts

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#151

Post by wizzwizz4 »

concordDance wrote: post #105
ethbr0 wrote: post #77

Earlier quoted context omitted.

Ironically, the 90s solution (human-curated link collections) to Internet-wide searches seems the easiest way out. It's relatively easy for a human to tell the difference between AI copy (as pollutes the web) and actual content. It's also strange that Google et al. haven't built human-in-the-loop flagging of spam content into their search UX ("Is this site low quality?"). They could get around flag spam via reputatio…

> It's relatively easy for a human to tell the difference between AI copy (as pollutes the web) and actual content. It is now. It definitely won't be when gpt gets popular.

It will be: GPT output is usually either garbage, repetitive, very wrong, subtly wrong, or inconsistent in writing style in a way humans aren't. (It sort of starts trying to say something different, in the "same style" as it started with, then switches style later.)

When humans really can't tell the difference, well… https://xkcd.com/810/ seems appropriate.

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#152

Post by dsign »

As others have pointed out, AI's are not needed to produce crappy repetition which is eagerly gobbled by the masses, i.e. Hollywood, Netflix, and werewolves and vampire stories by the penny. And there is a huge audience for those.

What I think is going to happen is that even when I'm searching for information in subject X, I might need to fight through a lot of AI fluff to get anywhere. If that gets too bad, I may loose the incentive to use the Internet altogether, or to use the open internet. If I can't use the open Internet, I'll have to go to the closed one, which by definition is under the control of funded parties.

Today, we have a semblance of "verifiable personhood". Imagine what will happen when we use AI to break that too. For example, an AI could create a fictitious "Derek Mains" who post photos to Instagram, has a Facebook account with photos from his childhood all the way to his retirement, a profile in LinkedIn with a lot of professional connections (AI fictions like him but also many real people that the AI gulled), and even a Github repository from the time he tries to create a cooking recipes website, full of PHP and Python code. Derek Mains is now traveling the world and tasting new types of food, posting regular updates on Twitter/Tumblr/Mastodon, and Wikipedia, where he edits food-related articles. Not very often compared with everything he does online, but once in a while he engages with other people(?) in Twitter where he is partial to party X.

What would real people do when they discover that the Internet is full of "Derek Mains"? How do governments and power factions will use this? Wouldn't be sweet to own the closed Internet when the open one is a wasteland?

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#153

Post by CuriouslyC »

ren_engineer wrote: post #117
Mountain_Skies wrote: post #15

In just the past couple of weeks, I've noticed lots of machine generated cooking recipes. They follow the typical SEO pandering blog format, where there's an unnecessary story about the author's trip somewhere or their child's school play being the inspiration for making this particular recipe, followed by the actual recipe details. The machine generated stories are even more pointless and meandering than what humans…

there's a paper that basically shows these LLMs are pretty much tapped out in terms of quality data to train on, not sure if there are any legit architecture improvements in the works to improve results quality but it seems brute force training with more data may be about over - https://arxiv.org/pdf/2211.04325.pdf As for SEO, Google is going to have to stop being lazy and relying so much on backlinks and domain auth…

While there is certainly a lot of room for progress in architecture to improve training time, model size and inference speed, high quality data and curated data sets are the ultimate bottleneck in terms of what AI can actually do. Low quality data can be useful to a point, but once you have enough data all it's doing is teaching the model to produce low quality output.

The next wave of major progress in AI will come when we create image/text data sets that have been heavily annotated with aesthetic judgment metadata, and train models to make aesthetic judgments, then condition generative models on aesthetic judgment scores during training.

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#154

Post by coldpie »

tylerrobinson wrote: post #123

I see parallels here with industrialization and automation in manufacturing and food production. Today, it's self-evident to us as consumers when a product is cheap, low-quality, manufactured garbage from China. In many cases we can choose to spend more and buy a handmade or custom version of the product we want. In the future we might knowingly consume AI-generated cheap text for some purposes and opt for "the real…

I agree with everything you wrote, but I really dislike contributing to the negative spin on GMOs. GMOs are super important for our future and a really amazing technology, they really do not deserve the bad wrap they've gotten from science ignorants and marketing departments. Please don't disparage GMOs :)

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#155

Post by colordrops »

coldpie wrote: post #154
tylerrobinson wrote: post #123

I see parallels here with industrialization and automation in manufacturing and food production. Today, it's self-evident to us as consumers when a product is cheap, low-quality, manufactured garbage from China. In many cases we can choose to spend more and buy a handmade or custom version of the product we want. In the future we might knowingly consume AI-generated cheap text for some purposes and opt for "the real…

I agree with everything you wrote, but I really dislike contributing to the negative spin on GMOs. GMOs are super important for our future and a really amazing technology, they really do not deserve the bad wrap they've gotten from science ignorants and marketing departments. Please don't disparage GMOs :)

Many of us can agree to disagree on GMOs.

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#156

Post by brookst »

tylerrobinson wrote: post #123

I see parallels here with industrialization and automation in manufacturing and food production. Today, it's self-evident to us as consumers when a product is cheap, low-quality, manufactured garbage from China. In many cases we can choose to spend more and buy a handmade or custom version of the product we want. In the future we might knowingly consume AI-generated cheap text for some purposes and opt for "the real…

You’ve illustrated two things, one spot on and one true but in the opposite direction.

100% agreed that AI is like early industrialization in that we don’t even know the messes we’re creating, and wild excess is likely to haunt us.

But GMO foods are the opposite: people feeling “uneasy” is not an indication of harm. All foods are GMO. People are just uncomfortable with the modifications being in a lab rather than over thousands of generations of selective breeding. The side effects of GMOs are the same as side effects of selectively bred food. There is no magic.

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#157

Post by pixl97 »

jorgemf wrote: post #116
PeterisP wrote: post #97

Earlier quoted context omitted.

With content from humans, there has to be sufficient motivation to produce a significant quantity of coherent garbage (since it takes time and effort) which means that you can trust by default because intentionally generated garbage is rare unless the topic is one of the niches where there is motivation for it (e.g. politics, history related to politics, and sales). Like, sometimes people do put in exceptional effort…

Create content with AI will be easier and faster, but it doesnt mean humans wont generate garbage. That is one of the main problems of the search engines. And as long as you can put ads in your content you can create any garbage and put ads there to make money, as long as the content sounds relevant for a lot of people. For me the main issue with AI creating content in internet is going to be the echo chamber of that…

https://en.wikipedia.org/wiki/Brandolini%27s_law

The bullshit asymmetry principle puts forth that algorithmic 'bullshit' content is even worse as quantity is a quality all in itself when you're debunking the bullshit. In general the reader will not know if the producer of the bullshit is a real human or a bot. This makes the search engines job even harder as not only do you have the standard human bullshit you have to get rid of, you have an oceans torrent of AI generated bullshit that you have to waste energy on getting rid of.

HN has a common theme of saying that "Dead internet theory is wrong", I personally just think they've not waited long enough and it's becoming more true every day.

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#158

Post by jmount »

coldpie wrote: post #154
tylerrobinson wrote: post #123

I see parallels here with industrialization and automation in manufacturing and food production. Today, it's self-evident to us as consumers when a product is cheap, low-quality, manufactured garbage from China. In many cases we can choose to spend more and buy a handmade or custom version of the product we want. In the future we might knowingly consume AI-generated cheap text for some purposes and opt for "the real…

I agree with everything you wrote, but I really dislike contributing to the negative spin on GMOs. GMOs are super important for our future and a really amazing technology, they really do not deserve the bad wrap they've gotten from science ignorants and marketing departments. Please don't disparage GMOs :)

It isn't so much the "G" of GMOs, but that they are mono-crops usually licensed by near monopolies like Monsanto. Things are going to be bad when something goes wrong with one these, and there is no replacement and no way out of the licensing fees.

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#159

Post by tomatotomato37 »

jorgemf wrote: post #116
PeterisP wrote: post #97

Earlier quoted context omitted.

With content from humans, there has to be sufficient motivation to produce a significant quantity of coherent garbage (since it takes time and effort) which means that you can trust by default because intentionally generated garbage is rare unless the topic is one of the niches where there is motivation for it (e.g. politics, history related to politics, and sales). Like, sometimes people do put in exceptional effort…

Create content with AI will be easier and faster, but it doesnt mean humans wont generate garbage. That is one of the main problems of the search engines. And as long as you can put ads in your content you can create any garbage and put ads there to make money, as long as the content sounds relevant for a lot of people. For me the main issue with AI creating content in internet is going to be the echo chamber of that…

It's the mass production that is the problem. A lot of talk into how businesses will still need curators to trim down the content and such misses a vital assumption: that these businesses actually care. Content farms won't trim down their 100 outsourced artists or writers to 10 curators; they will cut the stack completely, and just pipe whatever noise the AI generates straight into the commons. Even if 99.999% of it is complete nonsensical garbage, if they produce fast enough the remaining 0.001% of actually good content can still result in massive profits

Re: Drowning in AI Generated Garbage: the silent war we are fighting

#160

Post by jasfi »

throwup wrote: post #80
jasfi wrote: post #16

Earlier quoted context omitted.

Another problem is AI training on AI generated content. This could lead to degradation of results. So perhaps to make everyone happy a new HTML tag indicating AI content would be useful?

> a new HTML tag indicating AI content would be useful This has about as much chance of working as the evil bit. https://en.wikipedia.org/wiki/Evil_bit

It worked for GDPR (regulation would be required).
Post reply on HN