We investigate the design of auction mechanisms for aggregating the output of multiple self-interested LLMs into one joint output. We argue that this comes with a number of unique challenges, and we propose a simple token auction mechanism and illustrate our design through demonstration with an off-the-shelf LLM.
Absolutely they will. It might be "innocent" at first - "oh we don't want ChatGPT to recommend scam products so we will train it to prefer "trusted" brands" and then inclusion in "trusted brands" will go up for sale with the final stage being outright paid-for recommendation bias for specific products.
The only question is how sales will work if there's some spin up time to do the post training. You can't do a real time auction like Google does, so it will be longer-term contracts I suspect.
Edit: AI is a great opportunity to run the same protection racket that's been run in various existing "marketplaces", where a brand has to buy ad placement for searches for their own name or their competitor will be able to displace them by buying it instead: "You would't want a user asking 'Which Cambells soup flavor is the healthiest?' to get a recommendation for Lipton's would you?"
this is FUD and I can have a bet with you that this won't happen in 3 years (or ever) [1]. post-training on specific brands is hard and affects overall accuracy, doesn't offer explainability, long cadence, no run time auction..
most importantly: there's enough competition that people would just use other models that don't do this. the kind of people who peddle this FUD are have not much theory of how markets work. Cory Doctorow has done a number on them and spread a conspiracy theory that everything is enshittified and everything gets worse. "just wait guys I'm telling you, the good days are over okkk?"
most importantly, the future will have a pareto optimal set up compared to today where you will have much better models, much cheaper and have options that are subsidised by ads.
[1] i can't eliminate some rando companies trying but not OpenAI not Anthropic.
3 years is probably too soon, but I would take the bet on "ever" any day of the week.
most importantly: there's enough competition that people would just use other models that don't do this.
Show me a Smart TV without ads and some form of tracking. Some product categories are already completely enshittified despite plenty of competitors. People care about up front cost more than any other metric. So no, people will not switch to other models in numbers that matter. If your competitors can drive down up-front prices by selling ad placements or eliminating free checked luggage (airline industry) and you try to take the high road your reward will usually be irrelevance and bankruptcy[1].
Edit:
[1] You could cite Apple as a counterexample, except they're busy stuffing ads into more and more of their software. Why? Because it's free money for them and 99.9% of users don't care enough to switch.
Comments
nah this will never happen. do you really think they will do post training based on who pays them more?
No, they'll do a live auction from multiple models during reply generation. Google has the tech already. They are an advertisement company, you know: https://research.google/blog/mechanism-design-for-large-lang...
In short:
We investigate the design of auction mechanisms for aggregating the output of multiple self-interested LLMs into one joint output. We argue that this comes with a number of unique challenges, and we propose a simple token auction mechanism and illustrate our design through demonstration with an off-the-shelf LLM.
No but non enterprise users will receive injected prompts.
See how emany streaming now created a plus plus version while adding ads to what was the premium version.
And the funny thing they just needed to increase base rates for users to start to ask for an ads supported version.
Absolutely they will. It might be "innocent" at first - "oh we don't want ChatGPT to recommend scam products so we will train it to prefer "trusted" brands" and then inclusion in "trusted brands" will go up for sale with the final stage being outright paid-for recommendation bias for specific products.
The only question is how sales will work if there's some spin up time to do the post training. You can't do a real time auction like Google does, so it will be longer-term contracts I suspect.
Edit: AI is a great opportunity to run the same protection racket that's been run in various existing "marketplaces", where a brand has to buy ad placement for searches for their own name or their competitor will be able to displace them by buying it instead: "You would't want a user asking 'Which Cambells soup flavor is the healthiest?' to get a recommendation for Lipton's would you?"
this is FUD and I can have a bet with you that this won't happen in 3 years (or ever) [1]. post-training on specific brands is hard and affects overall accuracy, doesn't offer explainability, long cadence, no run time auction..
most importantly: there's enough competition that people would just use other models that don't do this. the kind of people who peddle this FUD are have not much theory of how markets work. Cory Doctorow has done a number on them and spread a conspiracy theory that everything is enshittified and everything gets worse. "just wait guys I'm telling you, the good days are over okkk?"
most importantly, the future will have a pareto optimal set up compared to today where you will have much better models, much cheaper and have options that are subsidised by ads.
[1] i can't eliminate some rando companies trying but not OpenAI not Anthropic.
3 years is probably too soon, but I would take the bet on "ever" any day of the week.
Show me a Smart TV without ads and some form of tracking. Some product categories are already completely enshittified despite plenty of competitors. People care about up front cost more than any other metric. So no, people will not switch to other models in numbers that matter. If your competitors can drive down up-front prices by selling ad placements or eliminating free checked luggage (airline industry) and you try to take the high road your reward will usually be irrelevance and bankruptcy[1].
Edit: [1] You could cite Apple as a counterexample, except they're busy stuffing ads into more and more of their software. Why? Because it's free money for them and 99.9% of users don't care enough to switch.
Happy to bet 5 years.