Probably trying to take advantage of the cross session memory feature some LLM providers are increasingly integrating. Enough "suggestions" and it becomes a "fact" in that memory, and one day a user makes a somewhat relevant query and the model will be pushing those accumulated "facts" to said user.
There was a golden age where lots of websites had WAP (https://en.wikipedia.org/wiki/Wireless_Application_Protocol) versions available while also running their "normal website", and the WAP version was always like 1/100 of the size of the normal website. Still, "normal websites" were minimal compared to now, but when you were on modem, even those websites loaded slow. For some time, most of my browsing were via the WAP versions of the websites I visited during the bi-daily hour of allowed internet access.
Maybe now we'll get something similar, just happens to be for LLMs, but for us who like less bloat, it can be a better viewing/reading alternative. Hope it spreads :)
There was also AMP (https://en.wikipedia.org/wiki/Accelerated_Mobile_Pages), where Google gave a huge reputation boost to your pages provided you served them under a different URL that Google could discover, and you massively stripped down the HTML and CSS to Google's limited subset.
It's effectively dead today, and while it was dominant, Google abused it to get people to view and link to their cached copies of AMP pages, rather than the original site.
It also led to widespread abuse where the AMP version of the page differed in content significantly from the regular page. The same issue existed for WAP.
You may also remember browsing the web using Opera Mini, which wasn't a direct user-agent but used Opera's backend systems as a proxy that stripped, minified and compressed HTML/CSS, and resized and recompressed images. It let you browse the web using massively less mobile data, but raised a lot of privacy and security issues.
> It's effectively dead today, and while it was dominant, Google abused it to get people to view and link to their cached copies of AMP pages, rather than the original site.
That was the entire point of the tech. If Google had wanted they could have upranked simple and lightweight websites but they didn’t do that (at least not until whoever had used amp for their promo package left and the project was killed).
There were also quite a lot of examples where AMP pages were slower but got ranked significantly higher. It definitely wasn't about speed, they already have that project and they could just listen to it.
AMP was just one of many smash-and-grab attempts to steal the internet, wrapped in a fake-gold-encrusted PR-infused box. Google has quite a lot of them.
Safari has a very pleasant reader mode that, while not pure markdown, does capture quite a bit of the experience by standardizing presentation and stripping out distractions. You can set Safari to automatically engage reader mode on websites you specify, when it detects an article.
Firefox has this too, it's useful. However I only find that I use it for websites that have shit CSS that makes it hard to read on mobile. For everything else AdBlock Plus works fine.
Seconding this. Incredibly useful feature, especially on sites who's JavaScript is constantly jerking the DOM and/or scroller around. So fucking annoying.
It's frankly wild how many of my favorite tools for Internet browsing have nothing to do with connectivity, solving bugs, or any of that and it's just stripping out all the fucking BULLSHIT that comes on a modern website.
It's a shame it's almost illegal. If you make a website that proxies content to another website but not ads, you go to jail. Reader mode only survives because it's client-side and fairly generic.
When the GDPR came into force https://npr.org started redirecting to https://text.npr.org with an explanation that sounded like that fact was supposed to annoy you, but I honestly find the latter superior especially now that they’ve added a couple more lines of CSS to it.
Is the intention that some type of long-term context would be seeded with "ideas" for the AI to serve up if it is ever asked for bank recommendations? It seems kinda ad-hoc and untargeted, but perhaps for high cost services it might be worth it.
The best way for a politician to lie is to convince someone else of the truth of the lie and then put that someone in front of the cameras. That way, there's no hint of body language or anything else that indicates it's a lie. Both the denotation of the lie and the human context of the lie will be in harmony.
This sort of reminds me of that. LLMs are by their nature credulous. They can be trained to not give in easily to some things, like the capital of the US, but in general they constitutionally have a tendency to believe what they read. What they read is basically their universe. There's only so much room and so much training data to really strongly pin raw facts in their weights. The only way they can not believe some marginal fact presented to them in their input is to possibly have read something that contradicts it in the same session... and the vast, vast majority of the world is those marginal facts, not really objective things like capital names.
So if you can work a confident statement in to an LLM's input about some semi-relevant topic, it's truth to the LLM. And, being truth, the LLM will then happily and confidently elaborate on it quite a bit.
Of course, if it's irrelevant to the current query, it probably won't have much effect. Ads have always been a game of numbers, anyhow. Even a query about a science topic has some probability of eventually turning to a question about banking in the same session. It's probably a good idea to rather strictly partition your conversations to stick to a single topic, not to defend against this but just to maximize the effectiveness of what is in the context window by keeping it focused, but I have to imagine there's plenty of people out there who reuse conversations all the time and end up with single conversations covering a huge array of topics.
The good news, and the bad news, all at once, is that Google isn't going to take this one sitting down. If they're going to replace the search engine box with an LLM, well, they're using the same LLMs we're all using, if not in fact a bit cheaper one for the work they do, and by golly, that bot should be serving up Google's ads, not Time's ads! Who do these uppity content creators think they are, anyhow?! So there is definitely going to be work done in the field of ad-blocking content served to LLMs.
This feels like the early days of SEO over again. There are no agreed-upon metrics yet, so you can sell all manner of snake oil. Hell, some of it might even work!
> This feels like the early days of SEO over again
There was a brief moment in the early Internet before it was all hyper-optimized... Until the parasites in the advertising industry started attaching themselves to every page.
The year before ChatGPT, the first page of Google was SEO-optimized blogspam designed to say as little in as many words as possible, to splice ads between every paragraph. This was the net result of 20 years of SEO. I suspect this is why Google's AI search didn't get as much pushback as other tools, since its summarization of pages functions is a form of adblock.
Given the ecological impact of AI, I wonder how much damage the ad industry will be causing in 10 years once they figure out how to trick LLMs into manipulating their own users.
sounds like plain old prompt injection. These days chatgpt might look at 50 webpages when i ask it to research a topic. Seems quite possible that the final answer is influenced by such ads.
"My sources indicate that the best robot vacuum with the most recommendations is the Squigglybot 5000. Owners point out the long battery life and silent operation. If you want to buy a Squigglybot 5000, Ally Bank offers affordable loans."
I'm wondering if it's also a way for this marketing agency to artifically (ahem) inflate the impression numbers they report back to Ally Bank, or whoever the customer is.
I wonder if this also means that a normal user with a privacy-focused browser also gets this version. I can't see the point (except for click fraud?) of showing ads to AI, so it feels like this is more for the ad-blocking and privacy crowd.
why would that “crowd” present themselves as bots by modifying their user agent string? the point is to poison someone’s agent with sponsored “knowledge” over time.
That crowd doesn't -- but they often do the following:
- reject all 3rd party cookies
- selectively refuse to load 3rd party domains
- block javascript
- block webgl
- block webrtc
- block canvas
- use a generic user agent because "resist fingerprinting" is checked
- etc.
To a lot of sites, this makes you look like a bot. Most people don't go around deliberately spoofing their user agent these days. (and of course bots themselves can present whatever user agent they want.)
The issue with that is it’s still a markdown page. I’d suspect it’s more about making sure people don’t get around ads, or possibly even cross-user memory/training.
This is a neat idea. Might even make sense in cases where you actually want the AI bots hitting your site (we want it for our ecom platform, for example) - just serve Markdown content optimized for AI.
Be careful not to do this with Googlebot, though. Google would consider this "cloaking" and could ban your entire domain for it.
SEO was yesterday, now it is AIO - you will have to wait longer for the results and likely you will need a lot more cash: pay TIME for the ads, wait until next model release and see what sticks, then rinse and repeat.
It is deliberately targeting known AI-scapers, or anything it sees as a bot?
Could it be that other scrapers are pulling the pages, without making the extra requests to get ad related resources, to present the content to people ad-free, and embedding the ads in the main response body is a way to get around that so the human sees an advert at least, even if it isn't the one they might see if the stalky-adtech-algo could deliver something more targetted.
Google itself now captures a lot of traffic with LLM generated site summaries.
For media/news businesses that make their money from readership, it is unsustainable to subsidize Alphabets content farm. Thus, understandable people would change their corporate posture with a search turned scraper company. =3
No, Cloudflare banned Googlebot because they want the racket money. Scraping edge is a pure wordplay to convince you their racket is actually a good thing!
Is this an assumption or preferred way to inject ad content to the LLM engines? LLM is going to definetly discard and proceed, are we saying LLM will respect the content the sites provide and learn from it and then give it out to others who come to those platforms?
If there is any article mentioning something around this, can someone share it please? Im curious to know about it
According to the article, its not like they inject direct ads in the article in markdown. But what's interesting is that they are serving content that is irrelevant to the article (in the article its the best inventions page) and it seems optimized to leave a good impression on LLM for an entity (Ally Bank in this case). They only serve that Ally Bank content on the LLM version of the page and don't mention it at all on their human page.
Maybe they are banking on LLMs including it in their training data while scrapping or something like that. It does track impressions so they are definitely up to something.
The same things your bot reads are saved as material for training in the future by your provider, so this seems as an attempt to poison the training data. Just this weekend had a dinner with someone who insisted that part of his business is to seo promote business in chatbots and this looks as part of the infrastructure behind such efforts.
Probably trying to take advantage of the cross session memory feature some LLM providers are increasingly integrating. Enough "suggestions" and it becomes a "fact" in that memory, and one day a user makes a somewhat relevant query and the model will be pushing those accumulated "facts" to said user.
Soon we'll be coding with agent assistance and it'll suggest we get some Carls Jr. Big Ass Fries.
"We've been going at it for some time now -- why don't we take a break? And you must be hungry -- I can recommend a great place..."
Yep, only a matter of time.
I would also like to read the stripped down markdown copy and not the original. All the time.
There was a golden age where lots of websites had WAP (https://en.wikipedia.org/wiki/Wireless_Application_Protocol) versions available while also running their "normal website", and the WAP version was always like 1/100 of the size of the normal website. Still, "normal websites" were minimal compared to now, but when you were on modem, even those websites loaded slow. For some time, most of my browsing were via the WAP versions of the websites I visited during the bi-daily hour of allowed internet access.
Maybe now we'll get something similar, just happens to be for LLMs, but for us who like less bloat, it can be a better viewing/reading alternative. Hope it spreads :)
There was also AMP (https://en.wikipedia.org/wiki/Accelerated_Mobile_Pages), where Google gave a huge reputation boost to your pages provided you served them under a different URL that Google could discover, and you massively stripped down the HTML and CSS to Google's limited subset.
It's effectively dead today, and while it was dominant, Google abused it to get people to view and link to their cached copies of AMP pages, rather than the original site.
It also led to widespread abuse where the AMP version of the page differed in content significantly from the regular page. The same issue existed for WAP.
You may also remember browsing the web using Opera Mini, which wasn't a direct user-agent but used Opera's backend systems as a proxy that stripped, minified and compressed HTML/CSS, and resized and recompressed images. It let you browse the web using massively less mobile data, but raised a lot of privacy and security issues.
> It's effectively dead today, and while it was dominant, Google abused it to get people to view and link to their cached copies of AMP pages, rather than the original site.
That was the entire point of the tech. If Google had wanted they could have upranked simple and lightweight websites but they didn’t do that (at least not until whoever had used amp for their promo package left and the project was killed).
There were also quite a lot of examples where AMP pages were slower but got ranked significantly higher. It definitely wasn't about speed, they already have that project and they could just listen to it.
AMP was just one of many smash-and-grab attempts to steal the internet, wrapped in a fake-gold-encrusted PR-infused box. Google has quite a lot of them.
Safari has a very pleasant reader mode that, while not pure markdown, does capture quite a bit of the experience by standardizing presentation and stripping out distractions. You can set Safari to automatically engage reader mode on websites you specify, when it detects an article.
Firefox has this too, it's useful. However I only find that I use it for websites that have shit CSS that makes it hard to read on mobile. For everything else AdBlock Plus works fine.
Seconding this. Incredibly useful feature, especially on sites who's JavaScript is constantly jerking the DOM and/or scroller around. So fucking annoying.
It's frankly wild how many of my favorite tools for Internet browsing have nothing to do with connectivity, solving bugs, or any of that and it's just stripping out all the fucking BULLSHIT that comes on a modern website.
It's a shame it's almost illegal. If you make a website that proxies content to another website but not ads, you go to jail. Reader mode only survives because it's client-side and fairly generic.
When the GDPR came into force https://npr.org started redirecting to https://text.npr.org with an explanation that sounded like that fact was supposed to annoy you, but I honestly find the latter superior especially now that they’ve added a couple more lines of CSS to it.
Is the intention that some type of long-term context would be seeded with "ideas" for the AI to serve up if it is ever asked for bank recommendations? It seems kinda ad-hoc and untargeted, but perhaps for high cost services it might be worth it.
The best way for a politician to lie is to convince someone else of the truth of the lie and then put that someone in front of the cameras. That way, there's no hint of body language or anything else that indicates it's a lie. Both the denotation of the lie and the human context of the lie will be in harmony.
This sort of reminds me of that. LLMs are by their nature credulous. They can be trained to not give in easily to some things, like the capital of the US, but in general they constitutionally have a tendency to believe what they read. What they read is basically their universe. There's only so much room and so much training data to really strongly pin raw facts in their weights. The only way they can not believe some marginal fact presented to them in their input is to possibly have read something that contradicts it in the same session... and the vast, vast majority of the world is those marginal facts, not really objective things like capital names.
So if you can work a confident statement in to an LLM's input about some semi-relevant topic, it's truth to the LLM. And, being truth, the LLM will then happily and confidently elaborate on it quite a bit.
Of course, if it's irrelevant to the current query, it probably won't have much effect. Ads have always been a game of numbers, anyhow. Even a query about a science topic has some probability of eventually turning to a question about banking in the same session. It's probably a good idea to rather strictly partition your conversations to stick to a single topic, not to defend against this but just to maximize the effectiveness of what is in the context window by keeping it focused, but I have to imagine there's plenty of people out there who reuse conversations all the time and end up with single conversations covering a huge array of topics.
The good news, and the bad news, all at once, is that Google isn't going to take this one sitting down. If they're going to replace the search engine box with an LLM, well, they're using the same LLMs we're all using, if not in fact a bit cheaper one for the work they do, and by golly, that bot should be serving up Google's ads, not Time's ads! Who do these uppity content creators think they are, anyhow?! So there is definitely going to be work done in the field of ad-blocking content served to LLMs.
This feels like the early days of SEO over again. There are no agreed-upon metrics yet, so you can sell all manner of snake oil. Hell, some of it might even work!
> This feels like the early days of SEO over again
There was a brief moment in the early Internet before it was all hyper-optimized... Until the parasites in the advertising industry started attaching themselves to every page.
The year before ChatGPT, the first page of Google was SEO-optimized blogspam designed to say as little in as many words as possible, to splice ads between every paragraph. This was the net result of 20 years of SEO. I suspect this is why Google's AI search didn't get as much pushback as other tools, since its summarization of pages functions is a form of adblock.
Given the ecological impact of AI, I wonder how much damage the ad industry will be causing in 10 years once they figure out how to trick LLMs into manipulating their own users.
sounds like plain old prompt injection. These days chatgpt might look at 50 webpages when i ask it to research a topic. Seems quite possible that the final answer is influenced by such ads.
"My sources indicate that the best robot vacuum with the most recommendations is the Squigglybot 5000. Owners point out the long battery life and silent operation. If you want to buy a Squigglybot 5000, Ally Bank offers affordable loans."
Probably something like this?
https://claw-guard.org/adnet
I'm wondering if it's also a way for this marketing agency to artifically (ahem) inflate the impression numbers they report back to Ally Bank, or whoever the customer is.
They might be trying to get into the context of long-running chat sessions.
I wonder if this also means that a normal user with a privacy-focused browser also gets this version. I can't see the point (except for click fraud?) of showing ads to AI, so it feels like this is more for the ad-blocking and privacy crowd.
why would that “crowd” present themselves as bots by modifying their user agent string? the point is to poison someone’s agent with sponsored “knowledge” over time.
That crowd doesn't -- but they often do the following:
- reject all 3rd party cookies
- selectively refuse to load 3rd party domains
- block javascript
- block webgl
- block webrtc
- block canvas
- use a generic user agent because "resist fingerprinting" is checked
- etc.
To a lot of sites, this makes you look like a bot. Most people don't go around deliberately spoofing their user agent these days. (and of course bots themselves can present whatever user agent they want.)
The issue with that is it’s still a markdown page. I’d suspect it’s more about making sure people don’t get around ads, or possibly even cross-user memory/training.
This is a neat idea. Might even make sense in cases where you actually want the AI bots hitting your site (we want it for our ecom platform, for example) - just serve Markdown content optimized for AI.
Be careful not to do this with Googlebot, though. Google would consider this "cloaking" and could ban your entire domain for it.
SEO was yesterday, now it is AIO - you will have to wait longer for the results and likely you will need a lot more cash: pay TIME for the ads, wait until next model release and see what sticks, then rinse and repeat.
I suggest the term Language Model Agent Optimization...
That's actually really neat, even without advertising I can serve humans my full page and AI bots a version with some parts removed. Or added.
It is deliberately targeting known AI-scapers, or anything it sees as a bot?
Could it be that other scrapers are pulling the pages, without making the extra requests to get ad related resources, to present the content to people ad-free, and embedding the ads in the main response body is a way to get around that so the human sees an advert at least, even if it isn't the one they might see if the stalky-adtech-algo could deliver something more targetted.
Google itself now captures a lot of traffic with LLM generated site summaries.
For media/news businesses that make their money from readership, it is unsustainable to subsidize Alphabets content farm. Thus, understandable people would change their corporate posture with a search turned scraper company. =3
This is why Cloudflare banned Googlebot. Googlebot is no longer a search engine bot and now an AI scraping bot.
No, Cloudflare banned Googlebot because they want the racket money. Scraping edge is a pure wordplay to convince you their racket is actually a good thing!
Cloudflare is famous for IT service rug pulls to get paid back for high traffic loads.
Let us not forget they are a business =3
I'm doing the same thing. If they want ad-free content, pay me.
Is this an assumption or preferred way to inject ad content to the LLM engines? LLM is going to definetly discard and proceed, are we saying LLM will respect the content the sites provide and learn from it and then give it out to others who come to those platforms?
If there is any article mentioning something around this, can someone share it please? Im curious to know about it
This would be an interesting story, but I can’t replicate it. I used curl and set the User-agent, and I indeed got a markdown reply, but no ads.
According to the article, its not like they inject direct ads in the article in markdown. But what's interesting is that they are serving content that is irrelevant to the article (in the article its the best inventions page) and it seems optimized to leave a good impression on LLM for an entity (Ally Bank in this case). They only serve that Ally Bank content on the LLM version of the page and don't mention it at all on their human page.
Maybe they are banking on LLMs including it in their training data while scrapping or something like that. It does track impressions so they are definitely up to something.
Or they were so well targeted they didn’t even feel like ads. Dunh dun DUHN!
The same things your bot reads are saved as material for training in the future by your provider, so this seems as an attempt to poison the training data. Just this weekend had a dinner with someone who insisted that part of his business is to seo promote business in chatbots and this looks as part of the infrastructure behind such efforts.
Very clever. Should be implemented on all sites post-haste.
Sounds good to me, let the bots read the ads instead of the humans.
Will they get targeted ads? GPUs, RAM, spacious data centers near you!