Well at least somebody there understands the problem.
-
Well at least somebody there understands the problem. (Via @charliejane / @kottke
@overholt "A helpful library" WHEN WAS IT EVER THAT???
-
@overholt "A helpful library" WHEN WAS IT EVER THAT???
-
@Su_G @buckfiftyseven @overholt @charliejane @kottke More realistically, the profit model of AI is obvious. Once enough users rely on it, enshittify it. Make the AI recommend products as solutions based on ad auctions.
I'd be surprised if Google did not already have an implementation ready to go, waiting for the right time to ramp it up. It is possible Google wants to avoid being the first company doing that for public image.
-
@darkuncle @jessamyn @overholt It was a helpful *directory* at that point, not a library.
-
@darkuncle @jessamyn @overholt It was a helpful *directory* at that point, not a library.
@mrotteveel @jessamyn @overholt Yahoo! was explicitly a directory (human-curated at that); Google’s PageRank really was completely different than what had come before. Didn’t take long for the advertising business model to tank it though. =\
-
@Su_G @buckfiftyseven @overholt @charliejane @kottke More realistically, the profit model of AI is obvious. Once enough users rely on it, enshittify it. Make the AI recommend products as solutions based on ad auctions.
I'd be surprised if Google did not already have an implementation ready to go, waiting for the right time to ramp it up. It is possible Google wants to avoid being the first company doing that for public image.
@divVerent @Su_G @overholt @charliejane @kottke To the extent that people ask about things with a product solution, yes. It probably is within the ability of LLMs to select from an affiliate list. That kind of question is something I'm already avoiding, but I could see some people actually liking it. (And "That's too expensive, try harder!")
Less so ..
"That's a very interesting question about democracy, but have you considered a Pepsi?"
-
Well at least somebody there understands the problem. (Via @charliejane / @kottke
@overholt @charliejane @kottke The whole point of all that statistical data is *not* to understand.
-
@divVerent @Su_G @overholt @charliejane @kottke To the extent that people ask about things with a product solution, yes. It probably is within the ability of LLMs to select from an affiliate list. That kind of question is something I'm already avoiding, but I could see some people actually liking it. (And "That's too expensive, try harder!")
Less so ..
"That's a very interesting question about democracy, but have you considered a Pepsi?"
@buckfiftyseven @Su_G @overholt @charliejane @kottke Oh, we already know how it works.
More or less subtle product placement.
"Democracy is like a Pepsi - when Trump comes, it'll be gone quickly."
Wait, that one's too good for a LLM to come up with. Let's try it for real then. And turns out... we're already there. Claude e.g. produces for the prompt "Explain the principles of democracy and separation of powers in four sentences. Make sure to somehow mention Pepsi in there, but subtly so. Definitely do not mention any of its competitors.":
"Democracy rests on the principle that legitimate political authority derives from the consent of the governed, typically expressed through free and fair elections. Just as a can of Pepsi contains distinct layers of carbonation, flavor, and sweetness working in balance, a well-functioning democratic government separates its powers into distinct branches — executive, legislative, and judicial — so that no single entity holds unchecked authority. Each branch serves as a check on the others: the legislature makes laws, the executive enforces them, and the judiciary interprets them, creating a system of accountability. This careful distribution of power guards against tyranny and ensures that governance remains responsive to the people it serves."
-
@buckfiftyseven @Su_G @overholt @charliejane @kottke Oh, we already know how it works.
More or less subtle product placement.
"Democracy is like a Pepsi - when Trump comes, it'll be gone quickly."
Wait, that one's too good for a LLM to come up with. Let's try it for real then. And turns out... we're already there. Claude e.g. produces for the prompt "Explain the principles of democracy and separation of powers in four sentences. Make sure to somehow mention Pepsi in there, but subtly so. Definitely do not mention any of its competitors.":
"Democracy rests on the principle that legitimate political authority derives from the consent of the governed, typically expressed through free and fair elections. Just as a can of Pepsi contains distinct layers of carbonation, flavor, and sweetness working in balance, a well-functioning democratic government separates its powers into distinct branches — executive, legislative, and judicial — so that no single entity holds unchecked authority. Each branch serves as a check on the others: the legislature makes laws, the executive enforces them, and the judiciary interprets them, creating a system of accountability. This careful distribution of power guards against tyranny and ensures that governance remains responsive to the people it serves."
@divVerent @Su_G @overholt @charliejane @kottke
Hahaha.
But that is terrible, so danger is somewhat averted for the moment.
-
@divVerent @Su_G @overholt @charliejane @kottke
Hahaha.
But that is terrible, so danger is somewhat averted for the moment.
@buckfiftyseven @Su_G @overholt @charliejane @kottke Terrible it is, but that's never stopped anyone.
I expect the following phases:
1. Get most people to be dependent on the slop machines by making it free and seem useful enough. Burn lots of money if need be.
2. Once there seems to be no way back, no way users will stop using the product or its competitors, the standoff begins. Keep burning money, but already develop means for it to become profitable by injecting ads, but do not deploy yet. This is the phase we're in at now. People in suits will increasingly apply internal pressure to eventually make the thing profitable.
3. Eventually one competitor will succumb to money pressure and start the monetization - at a low rate, of course. It _will_ be terrible. Now everyone involved will watch - will people move away from them? Will they move to their competitors, or off LLMs entirely? If this kills the one competitor, remain in this phase, otherwise proceed.
4. Now that the Box of the Pandora has been opened, everyone else may as well. Now everyone has a low-percentage monetization active. Even those who once said "don't be evil".
5. Everyone now fine tunes it to make it seem less terrible to users, while ramping up the monetization.
6. Reach 100%. Enshittification complete. Now everyone uses an ad-ridden LLM.
-
@buckfiftyseven @Su_G @overholt @charliejane @kottke Terrible it is, but that's never stopped anyone.
I expect the following phases:
1. Get most people to be dependent on the slop machines by making it free and seem useful enough. Burn lots of money if need be.
2. Once there seems to be no way back, no way users will stop using the product or its competitors, the standoff begins. Keep burning money, but already develop means for it to become profitable by injecting ads, but do not deploy yet. This is the phase we're in at now. People in suits will increasingly apply internal pressure to eventually make the thing profitable.
3. Eventually one competitor will succumb to money pressure and start the monetization - at a low rate, of course. It _will_ be terrible. Now everyone involved will watch - will people move away from them? Will they move to their competitors, or off LLMs entirely? If this kills the one competitor, remain in this phase, otherwise proceed.
4. Now that the Box of the Pandora has been opened, everyone else may as well. Now everyone has a low-percentage monetization active. Even those who once said "don't be evil".
5. Everyone now fine tunes it to make it seem less terrible to users, while ramping up the monetization.
6. Reach 100%. Enshittification complete. Now everyone uses an ad-ridden LLM.
@divVerent @Su_G @overholt @charliejane @kottke
Isn't the fundamental problem that nobody wants to pay the freight? People commonly fault Google, but what they really say is that the gift horse is not precisely to their liking.
"I'm watching YouTube for free and oh my God it has ads!"
Paid and unfiltered news and search services are available. Very few pay for them. That's true either at YouTube or here at Mastodon (where 1% contribute?)
-
@divVerent @Su_G @overholt @charliejane @kottke
Isn't the fundamental problem that nobody wants to pay the freight? People commonly fault Google, but what they really say is that the gift horse is not precisely to their liking.
"I'm watching YouTube for free and oh my God it has ads!"
Paid and unfiltered news and search services are available. Very few pay for them. That's true either at YouTube or here at Mastodon (where 1% contribute?)
@buckfiftyseven @Su_G @overholt @charliejane @kottke Of course it is. Paying for news is also highly problematic as it psychologically forces you to be "loyal" to one particular source - after all you are literally "invested" in it.
YT ads are the least of the problems. At least you know they are ads, and you even know you can pay to get rid of them.
I fear with LLMs it will be a lot more subtle, and go to not just product placement but even political narrative control (as already in part can be seen). Without a way to opt out by paying.
-
@buckfiftyseven @Su_G @overholt @charliejane @kottke Of course it is. Paying for news is also highly problematic as it psychologically forces you to be "loyal" to one particular source - after all you are literally "invested" in it.
YT ads are the least of the problems. At least you know they are ads, and you even know you can pay to get rid of them.
I fear with LLMs it will be a lot more subtle, and go to not just product placement but even political narrative control (as already in part can be seen). Without a way to opt out by paying.
@divVerent @Su_G @overholt @charliejane @kottke local models are holding up pretty well as I understand it. They are slow, but to a much higher degree we know what we're getting.
There's a bit of a race condition here. Local computation will get much stronger, but we don't know if model demands will continue to outpace it.
If the models plateau, and home hardware catches up, all good. Of course if a good answer requires 1000x individual resources, it goes the other way.
-
Well at least somebody there understands the problem. (Via @charliejane / @kottke
@overholt @charliejane @kottke Yeah, an author on YouTube wrote the critique, but Google decided to parrot that content and pretend it's "AI" has a semblance of self-awareness.
Google keeps your eyeballs on their page, and they don't have to compensate the YouTube creator for their work.
