3 ms·

The ads may not be announced. If ads can be subtly inserted “organically” through crafted weights then AI companies may try to claim that it isn’t advertising,

The ads may not be announced. If ads can be subtly inserted “organically” through crafted weights then AI companies may try to claim that it isn’t advertising, if it’s even possible to catch them doing this. For instance, advertisers could pay to have their product embedded as the “best” in a category during training. If this is done as a fine-tuning step then it could be re-run later as advertisers and base models change.

baby_souffle 10mo ago
How would the billing work for this? So much of advertising technology is tracking for the purposes of attribution. How does openAI know what to charge for a particular product and category? How do I know if my money was well spent to boost my product in that category? I don’t think you’re wrong! I’m just curious about how the new pricing models will work.
- yunwal 10mo ago
  > How would the billing work for this? So much of advertising technology is tracking for the purposes of attribution. This isn’t a necessary condition for an ad to exist. When companies pay for their name on a sports stadium, they use various proxies to tell whether their name recognition goes up, but by and large you just don’t know if it’s worth it.
- duchef 10mo ago
  Individually constructed models serving selected poisoned datasets. No different to adwords. If company bid is highest, customer is in selected demographic, topic is appropriate - answer query using biased model. It would be trivial to make a poisoned model that always rates the best duvet as DuvetCompany001 in all related queries for example. Then simply charge per impression.
  - iamnothere 10mo ago
    Yeah, this is what I was thinking. It’s not a PPC or PPI model, it’s more like you pay upfront to hopefully influence people over a longer period of time. It’s like brand placement in TV/film. Not clear if most advertisers would be interested, but I’m sure that some would be.

baby_souffle 10mo ago

How would the billing work for this? So much of advertising technology is tracking for the purposes of attribution. How does openAI know what to charge for a particular product and category? How do I know if my money was well spent to boost my product in that category? I don’t think you’re wrong! I’m just curious about how the new pricing models will work.

yunwal 10mo ago
> How would the billing work for this? So much of advertising technology is tracking for the purposes of attribution. This isn’t a necessary condition for an ad to exist. When companies pay for their name on a sports stadium, they use various proxies to tell whether their name recognition goes up, but by and large you just don’t know if it’s worth it.
duchef 10mo ago
Individually constructed models serving selected poisoned datasets. No different to adwords. If company bid is highest, customer is in selected demographic, topic is appropriate - answer query using biased model. It would be trivial to make a poisoned model that always rates the best duvet as DuvetCompany001 in all related queries for example. Then simply charge per impression.
- iamnothere 10mo ago
  Yeah, this is what I was thinking. It’s not a PPC or PPI model, it’s more like you pay upfront to hopefully influence people over a longer period of time. It’s like brand placement in TV/film. Not clear if most advertisers would be interested, but I’m sure that some would be.