3 ms·
> Therefore, there will be no more incentive to create those types of sites. Incentive is still there; scoundrels trying to make easy money on the Internet wil
by TeMPOraL 3y ago
> Therefore, there will be no more incentive to create those types of sites.
Incentive is still there; scoundrels trying to make easy money on the Internet will adapt. Before, they tailored their SEO spam to fool Google's search crawlers and ranking algorithms. Now, they'll tailor their spam to fool LLMs. There's a well-established adversarial approach for this too.
The LLM works as a discriminator, your job is to train/tune a generative model that fools the discriminator. If you can do it semi-reliably, you can put an API in front of it and sell it as "prompt injector as a service", and rake in the money. Couple more people do that, and all the ex-SEO scoundrels can get back to poisoning the commons for the highest bidder.
(Open source, like always, will come to "help", and let any small scamgency run their own "mix ads into content" models locally.)
The web will indeed change. But LLMs are like any other tool - they can be used for both good and for ill, and those using it for ill have much more money and drive. So I don't expect the change to be for the better; the tech is different, but the incentives and (severe deficiency of) ethics of the main players are still the same.
- mattlondon 3y agoThey can fool LLMs sure, but to what end? They need the traffic on their site to sell the ads - poisoning a LLM's response does not directly drive traffic to their individual site, so they won't profit. I don't think that some random SEO outfit has much more money than Google or OpenAI or Microsoft - they're not going to be able to "DoS" (as it were) a LLM to trick people to come to their site and click ads.