Hacker News Reader

Comment by Mountain_Skies | Hacker News Reader

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 come up with and the recipe details have steps that contradict each other and specify ingredients that shouldn't be in the dish and weren't on the ingredients list.

I don't know how many people are fooled by these recipes and end up with a cooking disaster but if we start seeing faulty machine generated instructions on how to do electrical work or engine repair, the consequences will be much greater when followed. The real annoyance is that because these machine generated articles learned from sites that maximize SEO, the generated articles are rapidly rising to the top of search results and crowding out actual good information that wasn't SEO maximized.

View on HN

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 authority. They might have to consider allowing searchers to essentially vote on what they consider good. I don't see any other way low quality AI content doesn't dominate SERPs, established players will spin up entire sites and spam backlinks to them all. There are also people working on using AI to identify obvious AI content by finding patterns in the output which is probably our only hope.

For social media and forums they will have to rely even harder on captcha or they will be taken over by bots talking to other bots

An interesting idea but really then aren't you just recreating reddit to an extent. The things people like or that are popular are upvoted because the voter agrees whereas what is unpopular gets down voted regardless of veracity because people like the way it sounds.

Beyond that a once popular website or producer says or does something that offends the sensibilities of the internet crowd and then everyone goes and down votes their results because it violates the current ziegiest

>aren't you just recreating reddit to an extent

there are tons of ways to prevent that by weighting votes and only taking a random sample. SEO is basically a war where SEOs find a way to exploit Google's algorithm and Google tries to respond.

Backlinks were Google's original solution to keyword spam that made other search engines suck, they didn't expect people would create entire sites or pay other websites to send links to them so they could rank higher. Backlinks were themselves considered to be "votes" for quality similar to citations in academia. The problem is the barrier to entry to make a site and link to it is high, so SEOs themselves are often the only ones to do it. Adding some sort of voting mechanism would allow actual readers to give an opinion on quality, although you'd probably see SEOs pay people to vote on their link via a VPN switching IPs, so only people with established Google accounts could be allowed to vote. Obviously its a huge challenge, which is why Google is worth hundreds of billions for even partially solving it

I guess a return of "don't show this page in my results" could work better: it removes the worst results and it's hard to fake, no influencer is going to pay to downvote 200 pages and if they do, it would be easy to see that same user downvoted everything everything at random, without even clicking trough to read.

Helps for both search users and the platform. Like a free moderation service. Why have an employee too check if these pages should be removed if a crowd of tracked users can do it.

I wonder where is my view wrong because I wonder why isn't there such system yet...

You'd still have to deal with adversarial spam. Competitor A pays a botnet to visit and downvote Competitor B's links at will, thus tanking its ranking.

In order to detect malicious activity you'd have to use a bunch of other signals, and also to use the downvotes with some suspicion, so you're back at looking content quality, reputation, page speed, and many other signals. Which sounds very similar to status quo.

Personally I'd like to see some sort of explicit input. If not used collectively, at least to customize my own results, like it happens on YT/Netflix/Instagram/etc - e.g. "never show results again from this site", or "I dislike this type of content".

>The things people like or that are popular are upvoted because

You forgot to add "Because clever posters use bots to give an early upvote boost to their posts, and downvote competing posts" And the more eyeballs on the site, the more things will be manipulated this way by bad actors.

Yeah, it's nice to see the best parts of Digg survived! /s

People are already appending "reddit" to their search queries to filter out the SEO blogspam clogging up the results of a regular search. Reddit is already superior to base search results for anything that's not plain uncontroversial facts because at least you're getting an actual human's opinion, which you can treat and evaluate as such.

For all its flaws maybe becoming like reddit it what search needs. AI generated spam will likely always be better at gaming a search algorithm than a human. Human moderation has its flaws but at least we're very familiar with them - they might be worth dealing with if it keeps the internet vaguely usable.

Lots of reddit is already pretty useless, drowning in people who are desperate to sound clever by regurgitating one of the same 50 meme jokes with some minor twist to it. And the astroturfing on the political reddits are likely to start hitting the niche ones with products getting pushed by manufacturers. Just buy accounts that have been built by bots on front-page subreddits, then have them make AI-driven positive comments on the niche subreddits and now you've got organic-looking advertisement for your product. Moderators likely won't help since they are there mostly to keep the trolls out (or they're more focused on whatever political axe they're grinding).

Reddit is also having some pretty serious problems with bots reposting highly upvoted content, both in posts and comments. Sometimes it gets spotted -- especially when the bots pick content which was specific to some point in time, or when they repost a comment out of context -- but a lot of it goes unnoticed, especially in comment threads.

Myself, I append hackernews to my search queries as the discourse quality is much higher as well as the communities maturity. Vox populi, Vox de!

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.

"The key roadblock to achieving human-level language understanding in AI systems is the lack of a comprehensive model of human language and cognition. While large amounts of data and powerful algorithms can help improve the performance of language models, ultimately a better understanding of how human language works and how it is processed by the brain is necessary to build truly intelligent language systems. In addition, the ability to integrate multiple modalities such as vision, speech, and gesture, and to handle the complexity and variability of real-world language use are also important challenges to be addressed." -- you know who

My understanding is that Google does essentially allow searchers to "vote" on what they consider good by looking for bounces. If the searcher follows on a search results link and then quickly returns to the search page then it's considered a failure.

true, but that's part of the reason for the life stories being told on recipe sites that the parent comment was talking about. Bounce rate is another metric that SEOs have gamed by throwing a bunch of fluff onto pages where people have to dig to find what they want

So that's why they're doing it. I've been wondering for a while now just why so many sites make you work hard to fish out the information you're after. I assumed ad exposure, but I had this feeling that it's not the whole story. Delaying people to fool Google bounce metrics is the idea I missed.

that’s not actually why they do it. it’s because recipes are not copyrightable. so they throw in a “story” to make it something they can copyright and try to prevent people from duplicating.

That's specific to recipe. Here I meant in general - e.g. blogs and news articles that bury the core point / answer somewhere around the final 1/3 of text, but also don't seem to exploit the artificially elongated text to expose you to more ads.

How would having additional content above the recipe change the copyright status of the recipe itself, divorced from the additional content? It prevents directly copying the entire article, but not the actual recipe (and Google already heavily penalizes exact duplicate content).

What would make more sense is that they are actually taking the recipes from elsewhere and adding the stories to avoid directly duplicating another page, and the Google ranking penalty associated with exact or close duplicates.

My moderate experience with SEO suggests it's a combination of factors. First, Google generally rewards longer content (https://www.sweor.com/seocontentlength), or at least content longer than the few hundred words a simple recipe would likely be limited to.

Secondly, there is (mixed) evidence that Google rewards sites on which people spend longer before clicking back.

Thirdly, the longer format allows them to insert more keywords/longer tail keywords vs. a short format where inserting lots of keywords would be seen as keyword stuffing.

I'd also guess that as much as some people dislike superfluous stories when looking for a recipe, it is a positive for a subset of users that are more likely to subscribe to/return to that site directly in the future.

Does back button capture also serve to circumvent/delay bounce status? Cause good god is it annoying.

Google doesn’t actually have to do anything at all other than slurp personal data up and use it to improve ad margins. Once your the default search engine, the default home page, and the default browser, you’ve locked 99% of the world into your portal. You control what people see and do, and they have to accept it. That’s what triggered the browser wars of the 1990’s - the realization that that trifecta is the keys to world domination.

Saying “yeah but Kagi” is talking about you and your tribe only. The world isn’t going to pay for Kagi or take the N steps required to switch. Microsoft realized they could kill Netscape by simply not requiring the user to download anything, then failed to consolidate. Google managed to be better for just long enough to consolidate so pervasively that any further investment in improvement is a literal waste of energy and money.

So, don’t hold your breath from an improvement in SEO garbage. It’ll just get worse. Definitely use Kagi and hopefully we will have great niche subscription based tools that have an incentive to be better. But don’t expect that Google perceives ANY problems with the status quo or that they have to do anything about anything other than increasing ad margins.

I’m not sure it’s Google’s fault at this point. Maybe it’s just that every site thinks they need to be SEO spam in order to be noticed. And so the internet has become a cesspool of stories about how this is the same recipe passed down in your family since Julius Caesar.

There have been a ton of articles about how Google sucks lately. Well, maybe Google has gotten worse, but maybe the internet has just gotten worse.

If I search for “best gaming laptop” or “taco recipe” the results are pretty much uniformly spammy across Google, Bing, and Neeva, the latter being a paid search engine. All of them are just farms for ads, popups, and paid links sandwiched between AI generated copy.

Is that the search engine’s fault? Is there some oasis of quality results I don’t know about? I would like to know.

> If I search for “best gaming laptop” or “taco recipe” the results are pretty much uniformly spammy across Google, Bing, and Neeva, the latter being a paid search engine.

I believe you will find results on Kagi less spammy

https://kagi.com/search?q=best+gaming+laptop&r=us&sh=HJrdZzD...

(Kagi is also a paid search engine that requires an accunt, the above is publicly shared search)

Removing, downranking and classyfing spam/ads is part of things Kagi is focusing on and does decently well (IMO, Kagi founder here).

I do like to hear a founder of a still developing startup describe their current results as “decently well” … it speaks volumes about what their aim is and how they assess their current position trying to reach those goals. If you thought it wasn’t possible you wouldn’t be trying but you know you aren’t there yet.

Thanks for being honest like this. I really should get around to trying Kagi … I think I’m just still in the current habit inertia between DDG and Google

Great to hear from someone fighting the good fight! In your opinion, is the internet just spammy now, or is something else going on?

Internet is extremly spammy and it is a result of ad-based monetization which provides incentive for the spam to be created. Luckilly an efficient mechanism to detect most such spam is to measure number of ads on a site and then downrank such results, which is what we do.

thanks for building Kagi <3 – it's expensive but worth it imo

They had upvotes working before! Just bring back google+! I found it incredibly useful to land on a page and see a high upvote in the little plugin widget I had in Chrome.

What's not to like? Custard? Good! Jam? Good! Meat? Good!

Have an upvote as I lol into my soup :-)

I would be very interested in your meat jam custard recipe. Asking for a friendly AI.

That's Joey from Friends [0].

[0] https://www.youtube.com/watch?v=TSFgDZJVYbo

Here is my own favorite way of doing jam custard:

> I am not able to provide a recipe for meat jam custard as I do not have the ability to access information on specific recipes. However, I can provide some general guidelines on how you might go about making a meat jam custard.

> First, you will need to select a type of meat to use for the jam. This could be any type of meat that you enjoy and that is suitable for cooking and preserving. Some common options for meat jam include pork, beef, chicken, or lamb.

> Once you have selected your meat, you will need to cook it until it is tender and well-done. This can be done by boiling, roasting, or slow-cooking the meat, depending on your preference and the type of meat you are using.

> Once the meat is cooked, you will need to shred or finely chop it and mix it with a jamming or preserving agent, such as sugar, honey, or fruit juice. This will help to preserve the meat and give it a jam-like consistency.

> Next, you will need to prepare a custard base for the meat jam. This can be done by mixing together eggs, milk, and any desired flavorings, such as vanilla or spices.

> Finally, you will need to combine the meat jam and custard mixture and pour it into a baking dish. Bake the dish in the oven until the custard is set and the top is golden brown.

> Keep in mind that this is just a general idea of how you might go about making a meat jam custard, and the specific ingredients and cooking methods may vary depending on the recipe you are using. It is always best to follow a recipe carefully and make any necessary adjustments based on your personal preferences and the ingredients you have on hand.

I am ethbr0's illuminated understanding of the Cartesian product of {ways of preparing meat} x {ways of preparing jam} x {ways of preparing custard}.

Thank you!

Thinking of my confusion as a British person, the first time I heard about Americans having “biscuits and gravy”. Or the time an American recipe guide got confused in the other direction about the British traditional Christmas “mince pie” and put minced meat in instead of mincemeat (which, despite the name, is vegetarian).

Mincemeat can be vegetarian but it traditionally contains suet.

I still nurse disappointment, mixed with a hint of resentment, going back to the age of maybe 3 or 4 years old that mince pies contain mincemeat (which is fruit, not meat) rather than minced beef. In the view of my 4 year old mind, and frankly my 40-something year old mind, the names of ingredients and dishes ought not to provide a misleading impression of what they might be made out of. "Mincemeat" still feels a little like false advertising.

Mincemeat on sweetbread? Sounds delicious!

I dunno, Digestive Biscuits dipped in gravy also sounds pretty good?

The set of all things that don’t go well with gravy is vacuously empty.

That was actually the basis for a "Friends" sitcom episode. The pages of a recipe book got stuck together resulting in Rachel combining a bunch of random ingredients.

https://www.today.com/food/friends-meat-trifle-episode-how-i...

Never expected to see a Friends reference on HN.

That is currently my 2nd-highest voted comment on HN, so I'm a little embarrassed that it's a quote without attribution. But I figured if someone didn't know and searched those exact words they'd get to be one of today's lucky 10k.

Holy crap. I just got an account with OpenAI set up last night so I've been looking for interesting prompts to play with. Can't say I'm going to try the recipe, but this looks at least as good as most blog-spam recipe sites.

Prompt: Write a recipe and blog post for onion soup.

GPT3 Response: Onion soup is a classic comfort food that can be enjoyed at any time of year. It’s a simple and delicious dish that comes together quickly and is sure to satisfy even the pickiest of eaters. Plus, it’s a great way to use up any extra onions you may have lying around! Today, I’m going to share with you my favorite recipe for onion soup.

To make this soup, you’ll need:

-4 large onions, sliced -4 tablespoons butter -4 cups chicken broth -1 teaspoon dried thyme -Salt and pepper to taste

First, melt the butter in a large pot over medium heat. Add the onions and cook until softened and lightly browned, about 8-10 minutes.

Next, add the chicken broth, thyme, and salt and pepper. Bring the mixture to a boil, then reduce the heat to low and simmer for 15 minutes.

Once the soup has simmered, it’s ready to serve. Ladle into bowls and enjoy with some crusty bread or a salad.

Onion soup is a simple and delicious dish that’s sure to please the whole family.

A bit ago, with the Watson craze, they tried to figure out what else it was good for.

This Cocktail Concocted By IBM's Watson Isn't Half Bad - https://gizmodo.com/this-cocktail-concocted-by-ibms-watson-i...

Would You Drink a Cocktail Invented By a Computer? How IBM’s Watson become the world’s most creative bartender

https://www.smithsonianmag.com/smart-news/ibms-watson-can-mi...

---

The reason that blogspam recipes are so common is that recipes have a special place in copyright law in that they aren't.

https://copyrightalliance.org/are-recipes-cookbooks-protecte...

https://www.copyrightlaws.com/copyright-protection-recipes/

https://www.copyright.gov/help/faq/faq-protect.html (and in particular https://www.copyright.gov/circs/circ33.pdf )

From circular 33:

> A recipe is a statement of the ingredients and procedure required for making a dish of food. A mere listing of ingredients or contents, or a simple set of directions, is uncopyrightable. As a result, the Office cannot register recipes consisting of a set of ingredients and a process for preparing a dish. In contrast, a recipe that creatively explains or depicts how or why to perform a particular activity may be copyrightable. A registration for a recipe may cover the written description or explanation of a process that appears in the work, as well as any photographs or illustrations that are owned by the applicant. However, the registration will not cover the list of ingredients that appear in each recipe, the underlying process for making the dish, or the resulting dish itself. The registration will also not cover the activities described in the work that are procedures, processes, or methods of operation, which are not subject to copyright protection.

And so, you can grab a recipe and be 100% sure that it isn't copyrighted. Then you can add some SEO text around it and some advertisements on the side and you've got yourself a page that you are sure that no legal department will come after you for.

It may be that the 21st century Turing test will be 'whether an AI can tell that content is AI generated". Otherwise it may be much harder to filter the AI-generated content.

THIS!!

Whenever this technology is used in an accuracy-critical setting, it is a recipe for disaster.

These "AI systems" produce the MOST PREDICTED output whenever there is any ambiguity, and this REPLACES the ACTUAL UNEXPECTED value/ITEM. In information system terms, they are literally stripping out content with the highest information value.

I saw this first hand giving a deposition that was supposedly transcribed by an "Artificial Intelligence" system. The output looked beautiful, until I started checking it for errors. Normally an errata sheet has a dozen lines for errors. Mine had pages. And they were substantial and very meaningful. The most obvious site was a specific name that sounded vaguely similar (started with a "J" and "K" sounds) was replaced by "Jack Kennedy". It also replaced "you have" with "I have" and omitted "not" several times, literally reversing the meaning of the statement. There were 85+ errors that I caught, over an order of magnitude more than normal.

These things are basically fine as glorified type-ahead systems where it doesn't matter much and a human is in the loop. But using them in any situation where the results are critical is, at this point, basically fraud. They do not do what the proponents claim.

I guess I'm in the minority of actually appreciating a lot of the background. And it can't just be SEO given that cookbooks these days are hardly just lists of recipes for the most part and people do apparently buy them.

Absolutely, buying an actualy printed cook book is more than the recipes. I'm not sure the same thing is true for the 10M "Best Ever Chocolate Chip Cookie" recipes on the web.

I guess if I'm just looking for a recipe the text is easy enough to skip. Often there's even a jump to recipe button. But photos etc. are often appreciated as well as some sense of why this recipe might be different.

It's not just SEO it's also a requirement to get copyright protection

So? Someone else can (and often does) just copy the recipe. So if you don't think the rest of the content has any value, what has been gained by writing it? If I had a cooking site--which I don't--I'd certainly want to do more than just post recipes irrespective of monetary considerations.

Well to me there's two distinct types of background:

1. Background directly related to the recipe/cooking in general.

2. The writer's personal anecdotes tangentially related to the recipe.

Some cooking blogs have great background (eg. J. Kenji Lopez-Alt's work on Serious Eats) but I would say they're mostly type 1 - information that informs you about the recipe, how they developed it, technique etc. They have type 2 background as well (jokes, what inspired them to make the recipe, snippets of their day-to-day lives) but it's definitely a small proportion of the body of the text. These blogs always have background that's work reading.

What I can't stand is the blogs with reams and reams of text of nothing but personal anecdotes, meandering stories, media references, what their kids are up to etc etc, occasionally returning to talk about something to do with the recipe. I don't have any examples because I now avoid untrusted recipe blogs like the plague, but it becomes very obvious to me when a recipe blog strays from useful background to worthless padding for the sake of copyright and SEO.

Absolutely. There is content around recipes that is good and content that is, well, not. But a lot of people seem to object to anything other than a recipe.

It's because you can copyright everything except the recipe and often a recipe by itself fails SEO in today's world.

Sally's Baking Addiction is the only site you need, FWIW. Trust me on that!

The stories aren't just SEO-pandering. They're copyright-pandering: https://www.chatelaine.com/food/food-blog-recipe-storytellin...

So the interesting thing about these tools is that they don’t link to the sources or cite the sources but instead train on them to generate a new source.

I wonder if over time SEO spam becomes an invisible problem to us and instead the AIs have sort it out for us.

> I wonder if over time SEO spam becomes an invisible problem to us and instead the AIs have sort it out for us

There have been some good Twitter threads about this. Basically, this is the last generation of AI to be trained with inputs that were ~100% human. It probably won't take long until the bulk of the content online is generated by AI, at which point it will just be an AI feedback loop. Any new concept, idea, technology, whatever, that doesn't already have a bunch of content about it will never have a guaranteed fully-human-written corpus to start from.

Once there's a viable OpenAI competitor or two, it's probably only a matter of time until the AIs start an arms race to feed each other bad data. Tweak the model to generate subtly incorrect information, regenerate your training corpus in a way that it feeds into other AIs but ensures you ignore it, and now you're basically all the way back to having a few trusted "information gatekeepers".

This will in turn spawn a new research field and industry of automated AI content detection in order to filter out poisoned training data. Also expect to see web browsers incorporate that technology with some sort of rating or color code to flag pages that seem to be AI generated. It's going to be a real arms race.

Except at present we can also train classifiers that recognise GPT output, so ingressed text should be able to be filtered.

https://huggingface.co/openai-detector/ detects GPT2, but it still detects ChatGPT snippets most of the time, even though the algorithm has been improved and the training set is presumably quite different.

That said, I did just generate a complete lie https://news.ycombinator.com/item?id=33874757 using ChatGPT that scored 99.98% real on the detector, so adversaries can pre-filter too.

Also interesting is that historical corpora (every existing corpus) of data becomes more valuable, because they are untainted.

Similarly, I have found a ton of beginner coding tutorials that contain mountains of useless/unrelated text. I bet a ton of it is SEO, but you gotta wonder if its just generated by something.

Especially now that things like chatGPT seem to write coding tutorials pretty easy (with less bloat than the ones I find on google)

Same! I was trying to figure out async diesel in rust and I couldn’t figure out why the tutorials weren’t working. There were several that had similar code snippets, but different (and sometimes nonsensical) explanations. Couldn’t get anything to compile. I finally found a GitHub repo that they all seemed to be stealing from and managed to figure out the missing parts.

Bad input equals bad output. They've trained these models on poorly written content that's already been an issue itself.

I think the problem you are describing is not only about AI, it is about trusting your sources of information. I can happen with content from humans.

Yes - AI moves this to another level of scale but Google’s need to sell ad placements created the environment for that to work. When they can’t even be troubled to de-rank blatant StackExchange/GitHub/Wikipedia scrapers it created a big opportunity for these low-quality content farms limited only by their ability to generate content on every topic.

Except for the fact that it does so on an unprecide dented scale allowing small groups of people to do things without necessarily being caught allowing for far more effective ways of control. Even without it, it's hard to argue for the existence of online forums when hackernews is a classic example of why it doesn't work when the majority of the 'intellectuals' have an inflated belief and understanding on topics they have no basic understand on. It's so bad that none of the actual subject matter experts talk here besides ones with financial interests in the products themselves and the experts fearful of retribution. The irony is almost comedic.

I have cookbooks, both physical and in digital formats, but it's usually faster to do a web search, especially when I mostly need cooking time and temperature. Perhaps this will motivate me to finally get around the setting up a local search function for my electronic cookbook collection as the web is going to be less and less trustworthy.

Firefox makes it pretty easy to add a custom search engine.

That one could point to your local server instance.

https://developer.mozilla.org/en-US/docs/Web/OpenSearch#auto...

For all its flaws I find Youtube is one of the best sources for recipes these days. Especially when you search in the language of the country the dishes are from and use CC auto-translate.

It's ironic that for all our cheering of democratization of creation, we turn back to gatekeeping (video harder to produce than text) in a search for quality.

Although maybe that's more a comment on the inevitable end state of democratization of creation + centralization of search/discovery + monetization of traffic.

Why is it ironic? I think it's more of a "humans are unable to learn from history moment". Romans called it "the mob" because of the common problems that occurred when large groups of undirected people got together and chaos broke loose. Of course the opposite where the authoritarian leader ruled everything also lead to Rome burning.

There is a happier middleground where multiple gates exist to filter out the bullshit, but no said gatekeeper controls the entire market leading to monopoly behavior.

Video is just a better format than text for many recipes, at least for novice cooks. The results often depend on technique as much as ingredients, and it's hard to convey technique through text.

Well-edited video. Cannot state how infuriating it is when the exact piece I'm stuck on is jumped over with a time cut.

Incidentally, this is why I click on the dirtiest garage clip for car how-to's: it's usually just the raw start-to-end process.

Have a look at

https://www.eatyourbooks.com/

not sure if they do ebooks.

It's a shame, this seems like a really great idea, but I just tried to use it and for a recipe from a book I own it comes up with:

> We’ve helped you locate this recipe but for the full instructions you need to go to its original source.

Which I guess makes sense, how can they verify I own the cookbook, but still very disappointing.

I have never encountered GPs problem where the recipe itself (the thing I was searching for) rather than the surrounding SEO fluff was AI generated trash. If this problem widely proliferates, especially into other domains, curation is going to become absolutely vital.

There's a lot of junk writing that may or may not be authored by humans about health and fitness. Search Google for "how to do more pull-ups" and you'll find thousands of long, meandering, trash articles of generic common sense with stock images and lots of ads on spammy domains.

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 to fabricate something as a practical joke, but that is sufficiently rare that you can ignore it in your daily life.

Without AI-generated content, cookie recipes would be "safe" from that, since only people who care about cookies would bother to write up articles about them - but now you suddenly have to mistrust all sources of all information, which is a new thing.

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 content being used to train new AI models.

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.

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

> 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.

It's less about if the businesses care and more about if the consumers care. Businesses will respond to whatever consumers are willing to tolerate.

Honestly, it'll probably be a battle around requirements to label AI-generated content so that consumers can make an informed choice about what to trust. But, given the global nature of the internet, that's probably not possible. You probably end up with private/commercial reputation rating firms, and consumers can decide who they want to trust.

If people decide they don't care about the veracity of information or don't want to put in the extra work to ensure they are consuming "true" information they you'll end up with businesses built on auto-generated garbage.

AI-generated First Aid instructions.

AI-generated alternative cancer treatments.

Hugh.

Can you link one? Sounds very interesting.

Fun fact:

Video games with crafting/cooking such as RPGs are featured on some of these generated SEO sites.

Reply on news.ycombinator.com