Junk websites filled w AI-generated text pulling in money from programmatic ads
technologyreview.com
technologyreview.com
[1] For the curious, "Thalagoya" for the terrestrial species, "Kabaragoya" for the swimming species
I'm talking about sites that just list all possible phone numbers in a range, for SEO purposes, with no useful information about each number, and then try to sell you access to their invasive databases of personal information.
The faucet turned off in 2020, when Amazon reduced affiliate payouts for many products [1] and cut off third-party affiliate networks [2]. There are still a lot of those sites around, but there's no longer the same level of incentives for them to be created.
[1]: https://www.cnbc.com/2020/04/14/amazon-slashes-commission-ra...
[2]: https://www.adexchanger.com/ecommerce-2/amazon-removes-third...
Upvotes/Downvotes never told me anything useful. The thumbmnail image, subject line and number of views are usually enough for me to know whether I want to give it a shot or not.
Honestly, given that ChatGPT produces higher quality output, the next generation of spam blogs will (sadly) probably be an improvement over the sea of crap we have now.
I think that's questionable. Gibberish is easily skipped over, bargain basement ESL writing with no familiarity with the product sticks out a mile, but with GPT you get a kind of reverse "uncanny valley" effect where it's written in polished English and says just enough in the opening paragraphs that sounds informed to lead you to read half the piece before you realise it's machine generated bullshit. So you waste a lot more time on something that has no more insight (and cost even less to produce)
So if demand stays the same, but the cost to produce it gets even lower, the ecosystem can support more players each pumping out their own version of all the spam sites. Stupid fake number example: Instead of spending $1000 a month to push out 100 articles, you spend $200 a month to push out 100 articles, and now only need to bring in $300 instead of $1100. So now there's another $800 worth of views that others could capture without putting you in the red, while letting them be in the green. So we may just end up with even more blogspam pushing the unique sources down the results list in Google. It's not likely to be meaningfully better based on my experience with "naive" prompting (the people creating these articles aren't going to be tuning their prompts to get good stuff per topic, they'll just get the quick and dirty stuff).
But I don't think that's where it's gonna get fun.
You know where there could be a lot of demand for higher-quality, harder-to-spot spam? Corporate product marketing and politics. Astroturfing in far more writing styles for far less money. Individuals running more fake accounts on Reddit, HN, etc, pushing more "unique"-but-repetitive copies of their views, while sprinkling in a fair bit of on topic info on a wider variety of other topics.
Hah, yep. I used to build tools to detect it at an adtech startup. A common approach at the time was to take someone else's text and use naive thesaurus replacement so that it would be just barely comprehensible but statistically look like english. So “You can catch the mouse” might become “you jar trap the rodent”. Glad to see how far technology has progressed! /s
There was some research a few years ago [2] into just how widespread this issue was in scientific publishing. The situation has likely only gotten worse with the introduction of higher-quality text generation LLMs.
https://twitter.com/janellecshane/status/1280915484351754240
Therefore, some form of reputation system remains necessary, like those used for scientific publications. I predict websites that provide a trusted reputation system will have a lot to gain in the future. Github stars, upvotes, retweets, citation count, or just good moderation - they will be essential in solving the spam/hallucination problem.
We haven't even solved the email spam problem, and that started 30 years ago! We have simply accepted that the largest players in the business (MS, Gmail, Yahoo) will decide on our behalf which emails we will actually see. If you want to start your own email service, fine, just keep in mind that the big players might think you're a spammer.
> to expect to use AI to solve a problem created by using AI, is wishful thinking
Fight fire with fire. Manual verification can't keep up with bots. We need "protective AI" to counter "aggressive AI".
Also, GMail is no longer good at detecting spam. It consistently puts my phone bill into spam, as well as other repeat emails I've marked as not spam. More importantly, I get bursts of actual, obvious spam in my inbox at least twice a month, possibly more often. Maybe Google has just given up and some other anti-spam system is better.
Mission accomplished?
Totally off topic, but I just think that the producers of shows like black mirror should just hire the guy and he’ll come up with plots for really disturbing things not the mellow obvious crap they try to push now…
It does sound like an improvement short term, but long term I wish there was another option.
Maybe even to an extent such that they will be actually useful
Do programmers in Google not use Google to search questions? Do they not have self respect seeing the bad quality product they’re throwing up on billions of people every day? I can’t imagine working like that.
Not really. Google has an excellent intranet search engine and mostly relies on in-house tools instead of commercial services (see https://github.com/jhuangtw/xg2xg). As a result its employees don't really need to rely on Stack Overflow, etc.
In the society I live in, self-respect doesn't pay the bills. I don't know where you're from.
Absent that, it sounds like the sort of excuse greedy people have always used to justify profitable harm.
Google has caused this problem by rewarding blogspam with high search rankings. Maybe they're trying to fix that, I don't know -- but if they are, they've already failed for at least a decade now.
I'd lay long odds Google stopped trying to rank pages themselves for those common searches long ago. Instead they look at what results users clicked on for a given search, and they are listed near the top. With a bit of randomness thrown in, so it can learn when new results become preferred. While saying it's a AI learning what to return for a search might be over the top, it's I doubt it's too far from the truth either. But you need a lot of people searching get the data to train the AI. Maybe Bing suffers from not having the users.
I'd also lay long odds that the search results are weighted by the ad revenue they generate for Google. Ie a site gets a boost if it is more likely to generate ad revenue for them if the user goes there. Competitors limit how far that can push that - they still have to be the best search engine. But as their primary goal is to earn money via ads they just need to be slightly better and use the remaining headroom to maximise ad revenue.
This is pure speculation on my part of course, and mostly evidence free. But I'm pretty sure Google engineers are smart enough to identify quite a bit of the SEO spam I see in the top of their search results - yet it remains there. This would be a quandary - if it wasn't in Google's interest for it to remain there because the spammy pages inevitably contain huge numbers of ads, and Google is at the end of the day an ad company that makes money from people clicking on ads.
There is information expressed/stored in the Web that is of relevance to the user's current information need. This is in pages, some of which are higher value than others. There is also a constant battle with bad actors trying to promote information or pages beyond its value or possibly relevance.
If interactive Web search is still going to exist at all (not just BS all our work with AI), then this is an opportunity for Google or someone else to step up.
This could mean aggressive new ideas that weren't viable before. For example, consider going full anti-BS. Basically, any whiff of attempting to manipulate search hits, or stealing/republishing content, or using generative AI is considered a strong signal for low quality, punishable by search engine death sentence. Suddenly, legacy genuine original content percolates to the top, and there's once again a market for high-value content creation. This one could be hard to do sufficiently well, and there might be conflict of interests in disrupting the status quo (e.g., some of a search/adtech company's revenue streams and departments/careers might currently depend on delivering low-value search hits), which could make it painful or face internal resistance.
I don't know who has the will, resources, and agility to do this. One existing player that could've done it with startup thinking a couple decades ago seems to have become more a machine and careers-driven for a long time. Another player has spent decades often operating in bad faith towards everyone, and though they've shown many times that they continue to be capable of long-cons, I'd guess they're probably culturally incapable of reasoning holistically about an approach fundamentally about integrity. Most potential upstarts don't have the know-how nor resources, and startup thinking would need to get back out of investment scam mindsets (whether VC-growth, crypto scams, or founders just trying to get their exit) and go back to trying to do something genuinely great for its own sake.
We are doing something like this at Kagi search.
While it is difficult to detect each of these on their own, what is easy to detect is the purpose of the web page if it has advertising on it (and most of these pages exist for the purpose of monetizing low quality content). So we downrank ad-filled pages in our results. It is a simple but effective method for filtering out low quality content.
What I've come to realize is that most sites that actually get significant traffic these days that run ads (so, all of them) don't really care too much if companies paying for ad space get a good ROI. There are so many stories of companies launching a campaign on Facebook with something like a $10k budget, not configuring the targeting exactly correctly, and burning through their entire ad budget in a few hours with only a few conversions to show for it. Looking at Twitter recently, the only advertisers they seem to be able to attract look to be overseas companies selling junk products for cheap, usually with some kind of "made you look" aspect like bikinis. On that last point, YouTube Shorts also seem to very heavily weight things that appear as if they're going to show some kind of nudity, which of course they don't, just very clickbait-y looking thumbnails in general that don't match the actual content of the short. Does YT Reels require that the thumbnail be a frame from the actual reel? That always seemed like the best policy to me, but of course people game it by putting one "hero" thumbnail frame in that doesn't actually come from the rest of the video to be able to select that as the thumb.
I know it's very much not in their interest to develop this since your eyeballs are the product but it would be so great if you could explicitly give the YouTube algorithm heuristics like: 'if the context is sailing exclude videos that heavily feature couples/relationships' or 'if the video sounds like that particular breed of shrill, hyperactive vlogger I'm definitely not interested'.
These are hard problems to solve. But google at scale efforts have failed.
I still have my popcorn ready for when the advertising bubble pops. There is no way web ads are worth what companies are paying for them now.
Ad spend, impressions, clickthroughs, and conversions are all heavily instrumented and tracked, the idea that there's a two-decades-long blind spot bubble strikes me as extremely farfetched.
Everyone knows there's a lot of fraud, its generally baked into the price at this point, and is why some sorts of ads are far cheaper than others, but (almost)[0] everyone also tracks their bottom line for returns.
[0] brand campaigns are different than direct response here, but that's an even older thing that predates the internet entirely so is unlikely to implode.
Definitely but I’d argue that they’re even more heavily gamed. The information asymmetry between the platform and the ad buyers is just too large. The incentive to misrepresent is basically limitless. Truly tracking effectiveness requires a degree of privacy invasion that even the platforms can no longer get away with.
Spamming more of cheaper ads for the same price does not count as ads getting better in my book. If anything, to me it indicates the opposite: if ads were getting better, campaigns would be getting smaller, shorter, cheaper.
I can't find any information about this in particular because I don't know the term to search for. Does anyone if this is true or how to find out more about this underground network?
No longer do they need to hire an army of cheap copywriters in some third world country to produce their spam. Anyone can run a spam farm nearly for free, and crowding the market is inevitably going to be driving down the profitability for everyone in the business as clicks and impressions are diluted. As is the trend with AI in general, it's really hard to build an effective moat.
We have now improved the A.I. to seem almost convincing, and sufficiently degraded the N.I.[1] such that convincing people takes much less effort.
[1] Natural Intelligence
Sadly, it seems that they are worth slightly more than what people are paying for them. I don't know who of the 5 billion online are motivated by these ads, but there are enough of them and they have enough money to make the system viable.
I have a friend who runs a local business entirely via Google ads, when he has space capacity he increases his ads spend and his capacity is used, when he has tight capacity he cuts back on his spend. He has a consumer facing business.
I don’t see an obvious catalyst for that in the short-term.
The status quo is that nobody knows how effective their ads are. The only people with the data necessary to even begin to analyze the effectiveness of ads are the ad-serving companies, who have a strongly vested interest in making them appear as miraculously effective as possible.
I mean if I Google a brand store the first result is usually and ad for the store and the next the store. If they measure the revenue from adclickers they hugely overestimate the value of the ad since the clickers where going there in the first place.
It is probably a complicit SC between Google and marketing folks. They are pimping their PowerPoints.
For example, the Microsoft IDE for dotnet is Visual Studio. You can get it by googling visual studio or you can go to visualstudio.com which redirects to visualstudio.microsoft.com
Now the typo squatted visual stuio dot com is interesting to me. It shows me a different scam depending on where I come from. Sometimes it is just a page full of ads. Sometimes, it is one of those “your registry has corrupted” kind of full screen scams, and sometimes it is just straight up pron. I’m saying all of this to dissuade you from visiting the website if you are on a potentially vulnerable machine.
The point is, someone could typosquat the New York Times or the Washington Post, or something like that and generate very confident sounding fake news now, right?
Maybe that’s what the tech giants want? They don’t want us typing the url but rather throw our hands up and say “{search engine} take the wheel”, and just type visual studio or nytimes and click the top result?
Almost no one in my social circle (except some IT people) types in urls anymore.
The red flag here wasn't the AI admitting it was "an AI language model". The red flag that this site was bullshit was the domain name CountyLocalNews.com
Everyone should be using an ad blocker, especially if going to random questionable websites, but the important thing that should have been hammered into people's heads a long time ago was that you can't trust everything you read online and that giving a few seconds consideration to your source is important.
We wouldn't have worry so much about things like AI spitting out fake news or deepfakes if people didn't treat @randomdude420 as a reliable source of information that should be taken seriously without question. @randomdude420, and I'm not picking on them if that's a real person, is just a random person online like me and should be treated as such. If something is important (and if it makes you angry or afraid it must be) verify it as best as you can.
Because Google has become trash we can't just click on the first few links that come back in search results and assume that we're getting accurate or quality answers. Since Google doesn't want to fix this, and nobody else is stepping up, people just have to be smarter.
It just sucks that it's getting harder and harder to find legitimate results at all since they're being drowned out by garbage.
> NewsGuard has a clever way to identify these junk AI-written websites...For example, one site called CountyLocalNews.com had messages like “Sorry, I cannot fulfill this prompt as it goes against ethical and moral principles …
`if str_contains(body, "Sorry, I cannot")` thanks NewsGuard, you've solved the problem pats NewsGuard on head while giving them an encouraging smile
It is a tendency to absorb whatever we read or hear - regardless of its information content.
IMO: Human languages evolved more for bonding than for conveying information. So, words exchanged were more to soothe the emotions of individuals involved rather than convey useful information about the state of their environment.
Hence we are fundamentally wired not to process the information content critically but to be soothed emotionally by whatever we read or hear
I can't speak to how bad things are elsewhere, but the average American only reads at a 6th grade level. That's also about where their understanding of science is. Their math skills are even worse. Americans today are not a good indicator of what humanity is capable of. We could always try to fix our educational system as opposed to sabotaging it. I have to believe we can do better.
So if google fixes this problem, their revenue goes down $13B/year. Seems like they will never solve this problem.
Which is bad... because I think the soul crushing work was what kept us from producing content at the rate a GPT assisted process will.
Today, search is horrible to use because Google has for years let these spam sites proliferate. Generative text content is the 1.5C event for text search engines -- it's too late to reverse it, we just have to find higher ground somehow.
[0] https://searchengineland.com/google-panda-update-guide-38110...
To be more precise, is doing that right now. I ofter hear (in my bubble) people replacing Google for ChatGPT.
There was all the stuff about singularity and how AI reinforces existing biases, but the most pressing problem seems to be “it’s seems almost right but also very wrong, way too cheaply”
Remember when Uber was the harbinger of a driverless taxi future? Now they're hoping to be profitable through food delivery, like an impoverished agrarian economy praying for rain.
Why not? This is the lowest-hanging fruit there is, and the open web's sole method of monetization has been ads.
Our voice are just getting better. Expect tons of spammy calls. Or you think you're calling a human tech support and it's just some machine blabbing about snake oil.
Sam Altman talks about having a 100 humans in the cloud working for you. It easily turns into a 100 humans in the cloud trying to juice every little penny out of you they can.
Google's algorithms are quite dumb, and LLMs are getting real good at parroting.
How can they even let them through?
The ad buy checks/credit card charges don't bounce?
Don't forget who is the customer in those transactions. Alphabet is an advertising company, not a search engine or video library.
Their goal is to maximize revenue, so if someone is willing to pay, why would they refuse to take the money?
Putting up a billboard with any of this crap would get someone in prison. And justly so.
>Putting up a billboard with any of this crap would get someone in prison. And justly so.
Sorry for the late reply.
By all means report the advertisers and Google to the authorities wherever you are. I certainly wouldn't be upset of some (or all) of those folks went to jail. In fact, I think it would quite nice.
I use an ad blocker so I don't see such things. As such, if you do please, by all means, involve the authorities.
But perhaps that’s why Google lets this harmful crap fester on their ad platforms. Perhaps they simply aren’t aware of it because everyone there runs an ad blocker. This would somehow be comforting, compared to the alternative.
I mean, there is no sensation here. Before AI-generated text, you could pay 5$ per 1000 characters of human-generated text, and it was used to do the same thing – playing Google's game.
The problem is that the feedback mechanism is far too easily compromised (both by intentional means and unintentional ones).
In order for such a scheme to function properly, the ad network has to be able to tell how many sales were made as a direct result of the click. The feedback loop needs to be nearly perfect. As things stand, it is terribly easy for ad purchasers to game, and not everybody is honest.
I agree that the concept is ideal, but I have yet to see an implementation that actually works without a great deal of additional effort (and a lot of human attention, and many many arguments…).
Protocols not platforms y'all. The game changed. Let them make money making nonsense. Cut through bullshit. Kill your masters. Host it yourself.
The most popular websites already run the maximum number of ads their users will tolerate. There is no scenario where websites “run more ads” to compensate for LLM bots.
Until it suffers the same fate…
Not to pick on the commentator, but at least on HN comment used it as an incorrect reference. I found it because I was going to use it as a reference, but I fact checked it and it was wrong. But also found a second ecosystem that fact check it as correct... I haven't looking into that ecosystem yet.
Incorrect information is good for SEO, because it's unique content. GPT can do it well with prodding. It's a very different beast to Markov chain or translation spam