ChatGPT is to begin showing users adverts for the first time unless they pay for a monthly subscription.
OpenAI, the bot’s developer, said it would start displaying adverts in the coming weeks based on what people ask ChatGPT.
However, it pledged not to sell users’ data and said the adverts would not influence what ChatGPT itself says.
Someone will, inevitably, start claiming what is denied there. That the company displyaing the ads sells our data to hte advertisers. Which isn’t what happens at all, of course.
Rather, the displaying company says to hte would be advertiser, well, we can slice and dice our audience these myriad ways. Male, female, age, likes to look at pages about the Roman Empire, has looked at ads for Italian property recently and so on. But they do not sell that data to the advertiser. That’s wholly an invention of those who would REEEE. Instead, the displaying company offers the advertiser the opportunity to advertise to those who meet those demographics. No one ever gets sent the information – they get allowed to use it to direct ads.
This whole idea that – say – Facebooks “sells our information” just isn’t true. But given how many people believe it it’s been a pretty effective lie now, hasn’t it?
Selling the data would make them less valuable to Facebook. So they wouldn’t, would they?
Exactly, they’ve got it exactly the wrong way around.
It’s not: the adverts influence what ChatGPT says.
It’s what ChatGPT says influences the adverts.
“You’ve been chatting about the influence of the Roman working class on the absence of Roman engineering literature, would you like to buy this spanner set?”
That’s a real thing, by the way. “Literature” was for the toffs. Nobody would lower themselves to write about something as plebian as bricklaying techniques or methods of concrete manufacture, so there is no extent literature about Roman bricklaying techniques or methods of concrete manufacture. It took until the 1800s to work out how Roman concrete was made, and to make similar quality.
Ummm.. Not really… The methods weren’t written down because they represented ways of plying a craft and controlling who could actually perform said craft.
Y’know… the economic incentives of Trade Secrets.
Still seen nowadays in the form of “Apprenticeships” and “Mandatory Union Memberships”, and “Certifications”, and ….. in a lot of countries.
Why ancient Greeks in dresses didn’t originate all that stuff gets named after them. It got generated whilst practicing crafts long before they were buggering their slave boys..
“Instead, the displaying company offers the advertiser the opportunity to advertise to those who meet those demographics.”
Which means the displaying company is profiting from the data of who its audience is and what they might want to buy. If it couldn’t tell the advertiser who is looking for a lawn mower, or who has a new baby, then it wouldn’t make any money from selling targeted ads would it? The fact that the advertiser doesn’t know that its Mr Smith of No 29 the High St who needs a new mower doesn’t mean Mr Smith’s data hasn’t been aggregated and then profited from.
Of course. The whole business model is to harvest and aggregate user data, then to sell advertising access to specific demographics. To sell the user data wholesale would be to destroy the business.
This of course is why, after buying a telly from Amazon, I get bombarded with more ads for tellies. One telly is never enough, is it? Clearly the model works.
Some stuff I buy again, consumables. Other stuff I’m not going to for a long time if ever. Who needs another computer monitor right away?
I guess it’s not worth separating the two out. Or there’s enough people unsatisfied with their initial purchase that they go shopping around later.
No customer data actually needs to change hands between OpenAI and the companies buying ads.
The idea is that e.g. “ABC Tools” buys an ad. They design it. Clicking on the ad takes you to a place where you can buy tools from them.
ChatGPT sees the stuff you’re looking at and inputting and on that basis shows you the ABC Tool ad.
If you click on it, then ABC Tool starts to get an idea of who you are, but not before then.
OpenAI does report that it showed the ad to e.g. 3056 people in January, so ABC Tool pays for 3 and a bit thousand impressions.
They probably slice up the customer base depending on what is being input into ChatGPT. So if you ask about the Roman Empire, then OpenAI may have determined that you’re more likely to buy tools and therefore it’s worth more to show you ABC’s ad than if you ask about plumbing. So it would then say showing the ad to people asking about RE is worth $0.01 per thousand, while showing the ad to people asking about plumbing is only worth $0.005 per thousand.
This has been happening for a long time; I think either Google or Facebook (or possibly even Yahoo) pioneered it.
Longer than Google/Facebook. When I opened a copy of Everyday Electronics in 1981, good god! there were adverts targetting me with electronics components supplies!
“If it couldn’t tell the advertiser who is looking for a lawn mower, or who has a new baby, then it wouldn’t make any money from selling targeted ads would it?”
The advertiser doesn’t need to be told who is looking for a lawnmower, he just needs to know that OpenAI, Facebook etc. will display their advert to people who are (according to an algorithm, likely to be) looking for a lawnmower.
“Profiting from” our data, yes. If you’re getting something without paying for it you’re not the customer, you’re the product. But “selling data”, no.
ChatGPT is f**ked then. Google have Gemini AI. Google are the masters at context-sensitive ads. Every company doing ads is setup with Google already. And if people can get normal results and “AI” results in the same place, it’s just easier.
I was thinking this about all this “robotaxi” stuff. Uber are going to win it, aren’t they? You’ve touched down at Zurich, what do you care about? It’s going to your hotel. Robotaxi might be cheaper but if you’re not in a robotaxi place, do you want to open up Waymo to realise they aren’t there, and now you have to open up Uber for a human driver? No. You want to type in Toblerone Hotel and see the options. How long, how much for each option.
They have taxi stands at airports for that reason. The taxis line up and you take the next taxi in line.
Do you know anything about the taxi company’s reputation? Probably not, you just want to get to your hotel.
Streaming services have some of the same problems. I see two modes for watching: either you want something in the background, or you want to watch a particular show.
If it’s the first, then you actually want a service, and anything will do.
If you want to watch a particular show, then the streaming service model is annoying. You have to pay them money every month for the chance to have something on? And then there’s a good chance it’s not on the ones you’ve paid for?
Imagine if you had to pay Uber/Waymo/Lyft a monthly fee…
One of the undervalued things about Uber is that it’s a brand that people have a long-term relationship with. I want an Uber because I can see the price upfront and I know the driver can’t take the piss.
Ditto I use CityCab because I’ve used them for 30 years and have built up extensive user-knowledge about them, so can base future consumer choices on that built-up knowledge; which is impossible with ply-for-hire.
Do you know anything about the taxi company’s reputation?
Which is exactly why ply-for-hire is more tightly regulated than pre-book. The consumer has no knowledge, all the knowledge is with the supplier, the consumer has no choice mechanism, the consumer has no blocking power of poor suppliers, so the random unrepeatable un-pre-identified supplier has to be regulated for consumer safety.
Uber/Waymo/Lyft are Just Another radio controlled pre-book private hire, and should be regulated exactly the same as Just Another radio controlled pre-book private hire, and ply-for-hire can stick their complaints over in that valley by Slice.
Ply-for-hire was killed off by cellphones for me. Long before Uber came along. As the train left Didcot, I would call minicabs and tell them I’d be in Swindon in 20 minutes. So I wasn’t waiting at Swindon for a minicab for more than a few minutes.
Uber made it extremely efficient because of how huge the operation is. You request a car, there is one near you in about 5 minutes. And that you get to Manchester or Paris and don’t have to know who the rip-off merchants are. And of course, satnavs. The Knowledge gives you an edge, but not much of one.
How is this any different to the days of magazines especially when it came to subscriptions?
The publishers know your address and no doubt offered enticements to do surveys and carried out further market research about who was buying their product.
Even within an activity they were able to segment the market further eg in the golf magazine market some mags would be aimed at the younger players and others at the older market. Same went for women’s magazines. They could even do some cross referencing if you took a couple of magazines, say golf and vinage cars.
Now they know a bit more about you but it’s still only about putting stuff in front of eyeballs.
If we want to get het up about something the two sided nature of the online advertising business has the potential to do us far more harm.
It’s different because magazines used to be profitable. ChatGPT is not profitable and will never be profitable:
It has told investors it does not expect to make a profit until the end of the decade as it projects spending more than a trillion dollars to build out its business
Ads won’t touch the sides of the money hole they’re digging, and won’t annoy people into the kind of mass takeup of paid subscriptions they would need to make this thing wash its face. But this is a good indicator of where the industry’s going – it’s fantastically unprofitable to have datacentres full of enterprise grade GPUs burning electricity so your Nan can edit photos of her cats. So they hope, by making the free version less compelling, people will sign up for the equivalent of a second broadband or mobile contract so your Nan can edit photos of her cats.
Like the gold rush of giant American media companies into streaming a few years ago, it’s going to run into the problem that consumers have no money and aren’t interested in signing up for dozens of subscriptions.
It’s different because magazines used to be profitable. ChatGPT is not profitable and will never be profitable:
It was a general comment about selling our data not a specific comment on ChatGPT.
ChatGPT may or may not become profitable selling ads, I don’t really care because if they don’t they’ll go bust and someone else will get to use the assets and I’ll find replacement service.
I do care, because these fuckers are ruining the whole IT industry. Never seen anything like it, even the dot com bubble was restrained compared to this.
Microsoft has now shoehorned Copilot into Notebook. Why? Fuck you, that’s why.
Anyway, yes, the silicon they’re spending hundreds of billions on won’t evaporate just because it’s Enron 2.0, but what’s the real value of a datacentre full of boxes that can cost $250K per server? We’re going to find out.
What they’re hoping to charge you £20/month forever for is probably going to be a boring included standard feature on Apple or Android phones within 5 years so you’re right. LLMs won’t go away, but the consequences of the financial crash they’re about to cause will be felt for decades. Or at least until the next financial bubble, where everybody invests in time travel companies, hoping to go back and recoup their AI losses.
What about that Chinese AI that can run on an old tin can? Don’t need no steeenkin’ Nvidia data centres for that.
They showed the barriers to entry to LLM aren’t as first thought. And also, there’s nothing to stop authoritarian governments just training their own LLMs and forcing citizens to use that.
My take on this is that LLMs are going to be specific. Like Adobe Firefly or Github Copilot. Smaller training sets and for people who will pay money for results and like an extension of existing functionality.
The options were
– the same AI for 10% of the hardware
– 10 times as much AI on the same hardware
The market chose the latter.
Download metapad.
Log on as Admin
Rename metapad.exe notepad.exe
Copy notepad.exe C:\Windows\System32\notepad.exe
job done.
Microsoft Six Seven
¯\_ (ツ)_/¯
I don’t get it 🙁
Sorry, my bad. Your comments sent me off looking at how Microslop is doing and it seemed to me that they might be busy killing themselves off. Couldn’t be arsed to write it up for a near dead thread, hence the meme.
Thing is, monetising your world dominating PC operating system and office apps into “AI” data harvesting and advertising delivery agents might sometimes be a good idea. But not when the very existence of the home and office PC itself is under threat by the various machinations of the “AI” bubble. If no-one can afford a PC, they might instead have a thin client that accesses an “AI” datacentre providing the number crunching via browser and broadband. No-one needs Windows FU for that.
Anyone beyond a regular Currys trog (who’ll choose a laptop because it’s got the nice Co-Pilot sticker on it), fucking hates “AI”, Microsoft, Google, Nvidia and all the other tech cunts.
Absolutely. It’s just a more sophisticated version of that. A more extreme demographics thing. Like you once would have put ads for Jags in a Golf magazine. Vogue would have ads for white wine.
What Gamecock finds ridiculous is that when he enters a search string in his browser, he gets a myriad of responses, including an “AI generated” cockup.
If the damn browsers would just give me what the hell I asked for, none of this AI generated crap would be needed. Allegedly, AI finds WHAT I ASKED FOR, then processes it with LLM, then spits out its rendering. If it just gave me what I asked for, it’d be done. But this begs the question, “If they can tease out specifically what I want, then process it, why can they just give men what I want?”
BTW, LLM is:
Hocus pocus fucking magic. With mirrors.
Just give me what I asked for. I know you’ve got it.
It’s a great advance in human – computer interface, and generative text, images, video and music is revolutionary just for the fact you can now tell a computer what to do in naturalistic human language, and it will mostly produce relevant output.
The smoke and mirrors are on the financial side and the sales pitch of this type of AI being able to hyperscale with massive compute, while improving business productivity in the now. None of those things are necessarily true.
The reason they added Gemini to Google Search is because Search has been broken for years. The new AI summary is often quite useful because it’s a contextually appropriate use case and so far hasn’t told me The Beatles recorded a song encouraging people to leave dogs in hot cars (hot car! whoo!). However, one of the reasons Google Search declined is due to the algorithm itself becoming less useful over time as SEO scumbags polluted the web with their bullshit. LLMs are highly dependent on scraping the Internet for training material, but the existence of LLMs accelerates the pollution of the Web with digital garbage… this is how you end up with evil robots kidnapping humans and forcing them to breed so they can consume our untainted raw human data.
But it’s also possible ChatGPT will get stupider with time, not smarterer.
Of course! The forces that corrupted browsers will also corrupt AI. E.g.:
Edit: It might come quickly. Difficulties monetizing the service may generate a flood of schemes to pay for it.
“However, one of the reasons Google Search declined is due to the algorithm itself becoming less useful over time as SEO scumbags polluted the web with their bullshit.”
If you put in “hotel wiltshire” into Google you mostly get guides to hotels in wiltshire. Because the guides have better SEO than the hotels do.
It’s why more and more I use specific search things. I don’t waste my time finding a hotel in Wiltshire on Google. I go to booking.com and look for it. Which also has ways to select criteria.
And Reddit is very useful. “Hey, who knows a good hotel in Leamington Spa” and you get some answers.
What’s wrong with people? If you don’t want your information sold, pay for the service. Why do people expect something for nothing? The service costs to supply so has to be paid for.
The Free Stuff Army battalions. Why I generally refuse to intact with people using freemail. I don’t want any of my supposedly private conversations perused by the service providers. If your e-mails are not worth paying for they’re not worth reading. You can fuck off.
A paid for service is no guarantee of data security or privacy. Any business is subject to takeover, changing its model to stay profitable, or just being cunty.
The one I use, it’s in the T&C’s.
I refer the honourable gentleman to my previous reply.
I only read letters that come hand delivered by a liveried footman and sealed with wax
“What’s wrong with people? If you don’t want your information sold, pay for the service. Why do people expect something for nothing? The service costs to supply so has to be paid for.”
It’s amazing how many people don’t get this.
Cory Doctorow bangs on about “enshittification” of online services, how they come out and are lovely and shiny at first, but the truth is that that the shiny version of it was always a lie, designed to build hype for investors and a customer base.