I'm positing a model where a third party does the influencing, not the company delivering the LLM/service. What's to say that it's an ad if the Wikipedia page for a product itself says that the product "establishes new standards for quality, technological leadership and operating excellence". (and no problem if the edit gets reverted, as long as it said that just at the moment company X crawled Wikipedia for the latest training round).
So more like SEO firms "helping you" move your rank on Google, than Google selling ads.
I'd imagine "undetectable to the LLM training orgs" might just be service with a higher fee.
How will these third party “LLM Optimization” (LLMO) services prove to their clients that their work has a meaningful impact on the results returned by things like ChatGPT?
With SEO, it’s pretty easy to see the results of your effort. You either show up on top for the right keywords or you don’t. With LLM’s there is no way to easily demonstrate impact, at least I’d think.
Disclosure is technically required, but in practice I see undisclosed ads on social media all the time. If the individual instance is small enough and dissipates into the ether fast enough, there is virtually no risk of enforcement.
Similarly, the black box AI models guarantee the owners can just shrug and say it's not their fault if the model suggests Wonderbread(r) for making toast 3.2% more frequently than other breads.
If Clorox fills their site with "helpful" articles that just happen to mention Clorox very frequently and some training set aggregator or unscrupulous AI company scrapes it without prior permission, does Clorox have any responsibility for the result? And when those model weights get used randomly, is it an advertisement according to the law? I think not.
Pay attention to the non-headline claims in the NYT lawsuit against OpenAI for whether or not anyone has any responsibility if their AI model starts mentioning your registered trademark without your permission. But on the other hand, what if you like that they mention your name frequently???
The point is that Clorox cannot pay OpenAI anything.
Marketing on your own site will have effects on an AI just like it will have an effect on a human reader. No disclosure is required because the context is explicit.
But the moment OpenAI wants to charge for Clorox to show up more often, then it needs to be disclosed when it shows up.
> But the moment OpenAI wants to charge for Clorox to show up more often, then it needs to be disclosed when it shows up.
Yes, I agree with this. But what about paying a 3rd party to include your drivel in a training set, and that 3rd party pays OpenAI to include the training set in some fine tuning exercise? Does that legally trigger the need for disclosure? You aren't directly creating advertisements, you are increasing the probability that some word appears near some other word.
With Google it's kind of ok as they mark them as ads and you can ignore them or in my case not see them as ublock stops them. You could perhaps have something similar with LLMs? Here's how to make bread.... [sponsored - maybe you could use Clorox®]
It's the same as it has been with all the other media consumed by advertising so far. Radio, television, newspapers, telephony, music, video. Ads metastasizing to Internet services are normal and expected progression of the disease.
At every point, there's always a rationalization like this available, that you can use to calm yourself down and embrace the suck. "They're marking it clearly". "Creators need to make money". "This is good for business, therefore Good for America, therefore good for me". "Some ads are real works of art, more interesting to watch than the actual programming". "How else would I know what to buy?".
The truth is, all those rationalizations are bullshit; you're being screwed over and actively fed poison, and there's nothing you can do about it except stop using the service - which quickly becomes extremely inconvenient to pretty much impossible. But since there's no one you could get angry at to get them to change things for the better, you can either adopt a "justification" like the above, or slowly boil inside.
Well as mentioned I don't even see Google's ads unless I deliberately turn the blocker off. I much prefer that to the content being subtly biased which you see in blogs, newspapers and the like.
So more like SEO firms "helping you" move your rank on Google, than Google selling ads.
I'd imagine "undetectable to the LLM training orgs" might just be service with a higher fee.