I feel like this is exaggerating a bit. Spam exists, but the internet is still very usable. It could certainly become worse in the future, but I find it pretty easy to sift through content these days to get what I need.
Also, I don't think the problem is necessarily AI for some of these complaints. Twitter replies are awful because you can pay for increased visibility. On a platform like Reddit where its popularity that determines visibility, the issue virtually disappears. Same issue for SEO spam websites, it's a function of Google's algorithm which incentives brainless keyword spam rather than it being an AI issue. Both of these issues predate generative AI.
The braindead children's youtube videos, for what it's worth, also predate AI.
I feel like these are just growing pains from a revolutionary new technology. Certainly the printing press can enable the spread of a lot of low quality content and misinformation, but we managed to work out the kinks.
> Certainly the printing press can enable the spread of a lot of low quality content and misinformation, but we managed to work out the kinks.
I think the author's point is that generative AI is a completely different animal than the printing press. The printing press could be used to spread both factual information and misinformation alike, and for various reasons factual information seems to have predominated, and high quality information is at least readily available, even if it's not the majority of printed work. Then there's the Internet, which can similarly be used to publish both information and misinformation. Perhaps due to the lower bar for entry and the speed of dissemination, the balance of information to misinformation and high-quality to low-quality content doesn't favor information or high-quality content as strongly as it does in the world of printed text, but at least the Internet always has the potential to spread factual information and high-quality content.
Then there's generative AI. Unlike the two communication technologies referenced above, AI can ONLY produce low-quality content. It is by design a statistical inference technique that generates content remixed from its training data. Without substantial human rework and rewriting, AI will always produce such low-quality drek as "it's hard to learn volleyball without a ball". And it's increasing promoted as a way to reduce human effort in writing, ensuring that people will continue to use it without supervising it's output, especially if they are trying to mass-produce content to make money. So now we have a new situation in which the majority of content produced going forwards is likely to be extremely low quality and perhaps contain substantial misinformation as well, whether intentionally or unintentionally. The author seems to posit that exposure to this type of content will negatively affect people's ability to learn to produce good, original content of their own, as they are not exposed to even passably good writing from a young age, so they cannot learn to emulate it.
If I filter out the AI spam and then the SEO spam, I'm not left with much anything. HN is one of the few online spaces where I feel like I can actually read the thoughts of actual human beings.
Also, like the other person commented, there's another generation coming up for whom AI generated content will be the norm.
Photoshop can't produce anything on its own; unlike AI, it needs a human to control each step of the process. AI generates statistically-likely text to follow a human prompt with no further intervention required by the human. And this lack of intervention means there is no human QC to ensure the statistically likely text meets any criteria for stylistic, semantic or logical correctness.
> And [AI is] increasingly promoted as a way to reduce human effort in writing, ensuring that people will continue to use it without supervising its output, especially if they are trying to mass-produce content to make money.
Yes, editing is still a thing, but all the crap content producers and SEO junkies jumping on the generative AI bandwagon and slathering AI-generated content across the Web can't possibly proofread all the content they're publishing - that would be counterproductive to their goal of producing as much content as fast as possible.
The problems he described have less to do with the technology and more to do with the fact that people just have way too much time on their hands. We consume garbage content because we don't have anything better that we want to do. Why else would someone read crap articles from SI written from a fake person?
Using the examples of both adults and toddlers to make his point is apt. At least the toddler has no choice.
(The author doesn’t say that. But he points out that pre-Altman OpenAI was afraid that GPT-2 was too dangerous to release, while now they let anyone use GPT-4 to spam the hell out of YouTube kids.)
I thought this was interesting, Erik is a great thinker and writer, but I felt as though the bad stuff he doesn't like here, specifically the weird videos for kids, might equally be blamed on photoshop, YouTube, or any other technology in the chain. Alternatively, it can be blamed on specific people and their specific actions.
Like, blame the person that makes the weird videos, not the technology that can both make weird/alien scripts for kids videos and also write my React components and describe what's outside my doorbell camera. It's some serious baby with the bathwater action.
General-purpose technologies are general-purpose.
The behavior he's upset about is currently human behavior. I don't know of anyone who has wired up GPT and asked it to make money and it has decided to start making YouTube videos.
At a certain point you have to acknowledge that if a technology enables harmful action on a scale previously unimaginable you have to start being very, very careful.
Machinists and gun makers are well aware how easy it is to make a machine gun or other illegal firearm, but they don't because that knowledge is locked behind experience, norms and expensive machines. This is why it was potentially a big deal when people started printing guns, not because it wasn't possible before to make dangerous or illegal firearms, but now it was an order of magnitude easier.
We have always had conmen, but running a good con was hard and you were limited to one at a time. Now with voice cloning and LLMs it is an order of magnitude easier, and infinitely more replicable.
I'm not sure the world is ready for the golden era of grifting, but I'm also not sure how you could ever stop it.
Just to add one thing: LLM’s as a force multiplier are much more than an order of magnitude (which is 10x). How long does it take a human to write a 1000-word article? Let’s say an hour, which is being generous. It won’t be a high-quality, well-researched article, but a human could probably get to that word count in an hour.
How long does it take an LLM to write a 1000-word article? Not 6 minutes —- that would be an order of magnitude faster. It would take ChatGPT maybe 10-20 seconds to spit out a 1000-word article. That’s 2 orders of magnitude faster.
I've been experimenting with making a Firefox plugin to hobble YouTube. I don't want to outright ban it (easily done using /etc/hosts) because there is good stuff on YouTube.
I talked to a guy at a conference who seemed really offended I might hack into my kids computers by modifying their YouTube viewing. But YouTube is hacking their brains right now and it drives me crazy. No one seems to be concerned about that.
I'm trying to get my kids to use their Ubuntu Linux machines instead of the fucking iPads. There is no way to inject myself into their YouTube experience on iOS devices. It's just them and Google, with no insight or control by me, their parent. I hate it. And I pay for YouTube ad-free. Why can't I have a dashboard to prevent my kids from seeing Ninja kids where they just shill junk from Walmart?
I mean that's a tough one, because the algorithm can be very subtle sometimes and trying to explain your reasoning as to why you think a certain trend in the content you're being suggested is an insidious form of behaviour modification is genuinely hard to do without sounding like a schizo who's spent too much time with Alex Jones.
Sometimes it seems to 'give up' and get frustrated - I literally was suggested some weird religious 'being gay is a sin' video followed by a literal vaginal speculum video once. Can't really misread that message.
> There is no way to inject myself into their YouTube experience on iOS devices.
Why don't you throttle YT on the iPads (via the firewall) so it's unwatchable, then blame Apple. You could make it somewhat intermittent to make it more convincing.
That's a good idea. Does that require pihole or something? I've got that installed but wasn't aware that I could do that. I'm not aware of how you can throttle traffic without intercepting on the device directly.
Does this require support from the internet provider? I don't see any options on the eero devices that come with Frontier. Does this require an intermediate router that supports throttling? I feel like this is something that pi-hole should support.
This is the dilemma. They aren't rational about what YouTube does to their brains. They don't understand. And YouTube isn't honest either. What's the right approach to fight it other than complete abstinence?
There are a lot of dishonest actors in the Internet world, how do you be honest in a sea of drug dealers in another industry?
Leaving kids to navigate conquering something that adults haven't isn't super appealing either.
With the lack of options out there, finding a way to preview content before it's made available seems the way to go.
Doing it early and keeping it that way seems to be the trick. But kids will often learn there is a "different" youtube on other family member's phones, or kids at school.
> I'm trying to get my kids to use their Ubuntu Linux machines instead of the fucking iPads.
Why? Like seriously. Why? What benefit does this give to your kid?
From your kid’s perspective, the laptop is going to be less useful, more hostile, it’s going to limit how they’re able to talk and relate to their friends.
Outside of your desires as a technologist and personal feelings about Linux, why should your kid want to use it? It feels like you’re projecting the ideals and values of your career onto your kids and then you’re surprised when they don’t share those ideals.
Good question. This is my opinion, but I feel I've got facts to back it up.
I don't want them using iPads because I consider the walled garden of apple to be hostile and predatory. Specifically, this is a place where I cannot control the device, where there are countless apps that are "free" that are advertised to my children without my consent, and then attempt to convince them to beg me constantly to purchase stupid coins or avatars or whatever. These app designers know that parents will give in after a hundred demands, and I resent that manipulation. And, I don't think Apple cares, in fact, I think Apple actively supports them for the 30% take they get.
That's a mixture of opinions and facts, so you'll need to sort it out.
It's ironic that Steve Jobs called the computer the bicycle of the mind. I doubt he would have called the iPad that. I'm hopeful my kids will learn a lot more about computers using Ubuntu. They definitely do explore much more on that then on their iPads which are purely consumption devices. My son was voluntarily using scratch to code a game tonight! He uses his iPad just for YouTube, so that feels like progress.
There’s a pretty wide suite of options built into iOS which lets parents restrict their child’s usage. It’s actually a problem they cared about a lot and spent many millions on, it’s a bit disingenuous to say Apple doesn’t put control in parent’s hands here.
Which parental control features exist in Ubuntu which are lacking in iOS?
My issue is that you can control time on iOS. But you can't control the individual apps and it is the wild west with every other app control, like YouTube. They offer controls but they are all completely different and it is exhausting to keep up with the different options. The kids have more energy to creep through the guardrails. I think that's by design, it isn't YouTube's business model to stop people from watching.
My kids are doing chess. After your comment, I searched for best YouTube chess channels. I love this one about amateurs (and trash talking while playing chess):
But, why does YouTube not make it easy for me to surface this content to my kids instead of these crappy channels like Ninja Kids or Aphmau which celebrate monoculture, consumerism, and fake accomplishments so they can sell junk from Walmart?
that's what the YouTube Kids app is for. you log in with your Google account and you create the kids accounts. you can select content by age and blacklist channels or you can whitelist only. Whitelisted only is probably the only way you can block some of the more annoying channels like Vlad and Nicki, they seem to have literally dozens of channels all full of the same drivel.
additionally, iPads have strong parental controls. you can limit time overall on the iPad and you can limit time on specific apps.
My coworker occasionally talks about his 4yo son watching youtube and it used to make me very uncomfortable. As someone without kids, I don't want to tell him what to do, but I thought it was a terrible idea.
Now, apparently his son has been watching lego and minecraft redstone tutorials, asking his parents to help him read the descriptions and linked text tutorials so he can build them! Youtube can really be an awesome place, but it's a double-edged sword for sure.
Speaking from experience, I was handed an iPod touch in middle school, and have been addicted to youtube ever since. Although most, if not all, of the passions that define me today were discovered/cultivated on youtube, all too often I'd never engage with them outside youtube. I really feel like I achieved nothing outside of class during high school and college, because I spent all my free time consuming.
What works for me now is using the unhook firefox extension [1] on my mac and android tablet, and then brave browser on my iPhone, signed out and with search history off. Basically eliminating all recommendations, I can still go to youtube with a specific purpose, but can't use it to doom scroll (that's what hn is for). I'd be happy to pay for premium, but not without a way to turn off all recommendations outside of search.
I'd be super interested to hear more details about the firefox plugin you're working on, is it open source?
I haven't open sourced it. I will do that soon. I'm very interested in your story, thanks for sharing! If you email me at chris@extrastatic.com I'll happily share the firefox extension when it is ready.
Youtube Shorts, Tiktok, Instagram, Twitter... basically any 'infinite feed' of content is no different from channel surfing of yore, which I think is also harmful to our brain and attention span.
This is a spin on the old debate of "guns kill people vs. people kill people". Both can be right, which is why effective regulation is necessary.
No such limits seem to apply to ChatGPT style word vomit generators, so the open web may literally turn into a no-man's land(fill).
The way to combat this is to zero in on small communities instead of large ones that attract spammers.
- Trusted blogrolls, webrings, OPML feeds
- Discourse-style messageboards over bot-ridden Discords.
- Smart filters on search engines that filter out blogspam sites (e.g. block if they contain social media "share" buttons).
- Return of apps like StumbleUpon for human curated websites instead of Pinterest-style sites that obfuscate original content and tries to add a "BUY ITEMS IN THIS IMAGE" overlay to every post.
Yup. This is going to fuel the return of the original Yahoo!-style human-curated internet search engine.
Aside from Google already being polluted to uselessness beyond redemption, the firehose of AI trash just makes it impossible, unless they are both motivated and able to get Gemini to discern between real content and trash (both unlikely).
Do you think there is a place for great art, content, generated by AI, artist using AI to generate and produce even greater art than what they normally produce ?
People keep posting these internet ruined by junk articles but my personal experience of it gets better and better over the years.
I think the paradox is because there is more of everything over time, quality content and junk. You can be selective and look at the quality stuff - it's not actually that hard.
64 comments
[ 2.4 ms ] story [ 126 ms ] threadAlso, I don't think the problem is necessarily AI for some of these complaints. Twitter replies are awful because you can pay for increased visibility. On a platform like Reddit where its popularity that determines visibility, the issue virtually disappears. Same issue for SEO spam websites, it's a function of Google's algorithm which incentives brainless keyword spam rather than it being an AI issue. Both of these issues predate generative AI.
The braindead children's youtube videos, for what it's worth, also predate AI.
I feel like these are just growing pains from a revolutionary new technology. Certainly the printing press can enable the spread of a lot of low quality content and misinformation, but we managed to work out the kinks.
I think the author's point is that generative AI is a completely different animal than the printing press. The printing press could be used to spread both factual information and misinformation alike, and for various reasons factual information seems to have predominated, and high quality information is at least readily available, even if it's not the majority of printed work. Then there's the Internet, which can similarly be used to publish both information and misinformation. Perhaps due to the lower bar for entry and the speed of dissemination, the balance of information to misinformation and high-quality to low-quality content doesn't favor information or high-quality content as strongly as it does in the world of printed text, but at least the Internet always has the potential to spread factual information and high-quality content.
Then there's generative AI. Unlike the two communication technologies referenced above, AI can ONLY produce low-quality content. It is by design a statistical inference technique that generates content remixed from its training data. Without substantial human rework and rewriting, AI will always produce such low-quality drek as "it's hard to learn volleyball without a ball". And it's increasing promoted as a way to reduce human effort in writing, ensuring that people will continue to use it without supervising it's output, especially if they are trying to mass-produce content to make money. So now we have a new situation in which the majority of content produced going forwards is likely to be extremely low quality and perhaps contain substantial misinformation as well, whether intentionally or unintentionally. The author seems to posit that exposure to this type of content will negatively affect people's ability to learn to produce good, original content of their own, as they are not exposed to even passably good writing from a young age, so they cannot learn to emulate it.
We technically inclined people are ahead of the curve.
One people get a “feel” for what AI content looks like, they’ll be able to filter it out like the do all existing spam.
It’s just there for information.
Non AI text that doesn’t have that AI flavor will stand out.
Also, like the other person commented, there's another generation coming up for whom AI generated content will be the norm.
Stopped reading after this. This is like saying Photoshop can only produce low-quality content.
Does this possibly include proofreading, editing, etc? Or does the AI automagically upload its content as soon as a human inputs a prompt?
> And [AI is] increasingly promoted as a way to reduce human effort in writing, ensuring that people will continue to use it without supervising its output, especially if they are trying to mass-produce content to make money.
Yes, editing is still a thing, but all the crap content producers and SEO junkies jumping on the generative AI bandwagon and slathering AI-generated content across the Web can't possibly proofread all the content they're publishing - that would be counterproductive to their goal of producing as much content as fast as possible.
Using the examples of both adults and toddlers to make his point is apt. At least the toddler has no choice.
(The author doesn’t say that. But he points out that pre-Altman OpenAI was afraid that GPT-2 was too dangerous to release, while now they let anyone use GPT-4 to spam the hell out of YouTube kids.)
Like, blame the person that makes the weird videos, not the technology that can both make weird/alien scripts for kids videos and also write my React components and describe what's outside my doorbell camera. It's some serious baby with the bathwater action.
General-purpose technologies are general-purpose.
The behavior he's upset about is currently human behavior. I don't know of anyone who has wired up GPT and asked it to make money and it has decided to start making YouTube videos.
We have a school shooting something like once every two days in the US, but can't put children before gun ownership.
I don't know how we survive this.
Machinists and gun makers are well aware how easy it is to make a machine gun or other illegal firearm, but they don't because that knowledge is locked behind experience, norms and expensive machines. This is why it was potentially a big deal when people started printing guns, not because it wasn't possible before to make dangerous or illegal firearms, but now it was an order of magnitude easier.
We have always had conmen, but running a good con was hard and you were limited to one at a time. Now with voice cloning and LLMs it is an order of magnitude easier, and infinitely more replicable.
I'm not sure the world is ready for the golden era of grifting, but I'm also not sure how you could ever stop it.
How long does it take an LLM to write a 1000-word article? Not 6 minutes —- that would be an order of magnitude faster. It would take ChatGPT maybe 10-20 seconds to spit out a 1000-word article. That’s 2 orders of magnitude faster.
The problem is that generative AI is competing with clickbait and it turns out it does the job cheaper.
I talked to a guy at a conference who seemed really offended I might hack into my kids computers by modifying their YouTube viewing. But YouTube is hacking their brains right now and it drives me crazy. No one seems to be concerned about that.
I'm trying to get my kids to use their Ubuntu Linux machines instead of the fucking iPads. There is no way to inject myself into their YouTube experience on iOS devices. It's just them and Google, with no insight or control by me, their parent. I hate it. And I pay for YouTube ad-free. Why can't I have a dashboard to prevent my kids from seeing Ninja kids where they just shill junk from Walmart?
Take the pads away, daddy.
That's what being a daddy is about, preventing harm when you see it occurring right in front of your eyes.
Nobody's gonna do it apart from you.
But, your advice is 100% correct.
Sometimes it seems to 'give up' and get frustrated - I literally was suggested some weird religious 'being gay is a sin' video followed by a literal vaginal speculum video once. Can't really misread that message.
Why don't you throttle YT on the iPads (via the firewall) so it's unwatchable, then blame Apple. You could make it somewhat intermittent to make it more convincing.
There are a lot of dishonest actors in the Internet world, how do you be honest in a sea of drug dealers in another industry?
With the lack of options out there, finding a way to preview content before it's made available seems the way to go.
Doing it early and keeping it that way seems to be the trick. But kids will often learn there is a "different" youtube on other family member's phones, or kids at school.
Why? Like seriously. Why? What benefit does this give to your kid?
From your kid’s perspective, the laptop is going to be less useful, more hostile, it’s going to limit how they’re able to talk and relate to their friends.
Outside of your desires as a technologist and personal feelings about Linux, why should your kid want to use it? It feels like you’re projecting the ideals and values of your career onto your kids and then you’re surprised when they don’t share those ideals.
I don't want them using iPads because I consider the walled garden of apple to be hostile and predatory. Specifically, this is a place where I cannot control the device, where there are countless apps that are "free" that are advertised to my children without my consent, and then attempt to convince them to beg me constantly to purchase stupid coins or avatars or whatever. These app designers know that parents will give in after a hundred demands, and I resent that manipulation. And, I don't think Apple cares, in fact, I think Apple actively supports them for the 30% take they get.
That's a mixture of opinions and facts, so you'll need to sort it out.
It's ironic that Steve Jobs called the computer the bicycle of the mind. I doubt he would have called the iPad that. I'm hopeful my kids will learn a lot more about computers using Ubuntu. They definitely do explore much more on that then on their iPads which are purely consumption devices. My son was voluntarily using scratch to code a game tonight! He uses his iPad just for YouTube, so that feels like progress.
Not everything is an engineering problem.
FWIW, Jobs specifically said he would never give his kids an iPad because it's bad for their developing brains.
https://www.nytimes.com/2014/09/11/fashion/steve-jobs-apple-...
Which parental control features exist in Ubuntu which are lacking in iOS?
Corporate mobile device management (MDM) policies can do a lot more for both on apple and android. Maybe that's somewhere to look.
Now they play Minecraft and Hello Kitty which is a slightly less braindead activity than YouTube!
https://www.youtube.com/c/CoffeeChess
Found in this list:
https://blog.favoree.io/articles/best-chess-youtube-channels...
But, why does YouTube not make it easy for me to surface this content to my kids instead of these crappy channels like Ninja Kids or Aphmau which celebrate monoculture, consumerism, and fake accomplishments so they can sell junk from Walmart?
additionally, iPads have strong parental controls. you can limit time overall on the iPad and you can limit time on specific apps.
Now, apparently his son has been watching lego and minecraft redstone tutorials, asking his parents to help him read the descriptions and linked text tutorials so he can build them! Youtube can really be an awesome place, but it's a double-edged sword for sure.
Speaking from experience, I was handed an iPod touch in middle school, and have been addicted to youtube ever since. Although most, if not all, of the passions that define me today were discovered/cultivated on youtube, all too often I'd never engage with them outside youtube. I really feel like I achieved nothing outside of class during high school and college, because I spent all my free time consuming.
What works for me now is using the unhook firefox extension [1] on my mac and android tablet, and then brave browser on my iPhone, signed out and with search history off. Basically eliminating all recommendations, I can still go to youtube with a specific purpose, but can't use it to doom scroll (that's what hn is for). I'd be happy to pay for premium, but not without a way to turn off all recommendations outside of search.
I'd be super interested to hear more details about the firefox plugin you're working on, is it open source?
[1] https://addons.mozilla.org/en-US/firefox/addon/youtube-recom...
So much of social media is TPIR style gambling for dopamine and worthless knowledge.
https://www.youtubekids.com/
No such limits seem to apply to ChatGPT style word vomit generators, so the open web may literally turn into a no-man's land(fill).
The way to combat this is to zero in on small communities instead of large ones that attract spammers.
- Trusted blogrolls, webrings, OPML feeds
- Discourse-style messageboards over bot-ridden Discords.
- Smart filters on search engines that filter out blogspam sites (e.g. block if they contain social media "share" buttons).
- Return of apps like StumbleUpon for human curated websites instead of Pinterest-style sites that obfuscate original content and tries to add a "BUY ITEMS IN THIS IMAGE" overlay to every post.
Aside from Google already being polluted to uselessness beyond redemption, the firehose of AI trash just makes it impossible, unless they are both motivated and able to get Gemini to discern between real content and trash (both unlikely).
Plus ça change, plus c'est la même chose . . .
I'm a bit confused about this comparison. Is there something about Discourse that makes it immune to bots, where Discord is not?
(There's a bit of humor in that Discourse has social media "Share" buttons, which are apparently evil)
So tasty and healthy.
People keep posting these internet ruined by junk articles but my personal experience of it gets better and better over the years.
I think the paradox is because there is more of everything over time, quality content and junk. You can be selective and look at the quality stuff - it's not actually that hard.