[bambax]

At least in this configuration ads are separate from answers. I fear there will come a time where ads will be intermingled in the response tokens. In fact SEO people on SEO forums have been talking about how to inject products into training for many months; but at some point the model operators will offer this directly.

[kubanczyk]

> At least in this configuration ads are separate from answers.

As a frog would say: at least the heat is separate from water, for now.

[satvikpendem]

Funny that people still perpetuate the myth about boiling frogs; they absolutely jump out at a certain hot enough temperature regardless of how slowly it is heated.

[Razengan]

I was literally wondering this when trying to fall asleep yesterday

[markovs_gun]

I think it's impossible for these ads not to eventually get incorporated into responses. There is a class of people who treat an LLM as their super smart best friend who they talk to about everything. It's a level of intimacy and influence that no technology company has ever had before. Do you think that it's possible that a publicly traded (I know OpenAI isn't publicly traded at the moment but they are trying to IPO) company with a fiduciary duty to its shareholders to maximize returns wouldn't use this intimacy and influence to make money? It's not just an ad for a new toaster, it's your best friend who you trust about everything telling you that the toaster he's recommending is the world's best toaster, as selected by a being with superhuman intelligence. Obviously not everyone will react this way, but for people already undergoing AI psychosis or at the beginning stages it will just make things worse.

[disgruntledphd2]

> but at some point the model operators will offer this directly

This is illegal in basically all of the developed world.

[markovs_gun]

Does that matter? AI companies seem more concerned with scaling at any cost regardless of laws broken because once they're big enough they're too central to the economy to face any real consequences.

[disgruntledphd2]

Weird, I thought I'd replied here.

This is different as the victims of this crime would be large enterprises with lawyers on staff, rather than a diffuse agglomeration of all the world's writers.

[selicos]

A $1.5Bil settlement just went in on stolen content used for training. It's arguably many times larger in long term damages especially compared to investment into those who stole the content.

If an AI company can scrape in billions of investment and keep going vs Facebook style "cost of doing business" fines, they will make that bet. If OpenAI can inject paid content, advice, misinformation, etc and keep operating or earn revenue today, it's worth doing. There are no ethics beyond optics.

I'm not saying these AI company CEOs would sell their mothers for more runway cash. But also nobody has seen their mothers since the second Sunday in May.

[oneeyedpigeon]

Copyright infringement used to be illegal in the developed world, then we started building exceptions for AI. There's no reason to expect this won't happen for advertising laws too.

[pixlmint]

They'll simply copy Google's playbook. First, clearly highlighted as advertisement results, usually the first three, so just with muscle memory you'll scroll down right away. Then they get bigger. Then the highlighting gets removed. Then it's just a tiny badge saying "advertisement".

[nibbleyou]

This is something that I noticed after doing a Google search on a new machine without uBlock Origin installed. I was unable to identify that the first link was an advertisement to a competitor's product of the brand I looked up. I remember adverts used to be highlighted really well in Google searches earlier.

If I am unable to differentiate between an ad and a search result then neither can a lot of non technical users including my parents.

Can't wait to see ChatGPT ads

[drybjed]

Black Mirror - "Common People" already did this.

[quotrend]

This already has a name — people call it GEO/AEO. I do the legitimate end of it for my own site: structured data, clean factual pages, llms.txt, so that when an assistant answers a question in my domain it cites the source correctly. Right now the incentives are healthy, because the only way to "rank" in an answer is to be genuinely accurate and citable.

Paid placement inside answers breaks that alignment. Why invest in being citable when you can buy the token stream? Search took about fifteen years to fully make that trade. LLMs seem to be starting on it in year three.

[satvikpendem]

Interesting that this was flagged when it is about the exact phenomenon being discussed, as I was also about to mention AEO until I read this comment. It is really not very different from SEO in that if you want your site to be ranked, it should have clean information, and don't try to game the system as otherwise you'll be banned like Google has done for decades now.