Assuming you can't just directly filter it, and even if it's hard to run a high quality LLM locally for the time being, it should be feasible to run a smaller model on the returned output of large commercial models to identify and remove any advertising. Actually I wouldn't be surprised if you could plug a model for filtering malicious content from all web pages right into the local translation framework Firefox already has.
This, I had actually thought about the ads idea but then realized that local models can filter it out.
But then again, the ads models have worked out fairly well for google and meta when ublock origin on firefox and for a loong time on chrome too was literally just go to appstore, search ublock origin, click download
That was it
I am sure that we are gonna see similar innovations happen for these ads too but then again, I think that the key point here is that the ad models works right now when we have ublock origin, so even if we have your tool, maybe most people (would still be ignorant about it?)
I think this is what is basing the whole AI industry in but this cannot be the way they can fund trillions of $'s spent.
I think what would happen more likely is that they will use AI to create a detailed profile of you and see the code generated, where its deployed, see what username it was and what email, link all your emails to every platform and every connection. Not going to lie, but its a scary thought.
I think something scarier can happen if it can actively shadow-ban you like its doing but now from the whole web in a more "personalized way" Imagine if google chrome launches out a feature that blocks websites which can be harmful to you creating an AI generated block-list I guess giving reasons why...
I think AI context windows are going to be an issue here and compute required to do this at scale so I don't really know and I am just rambling right now.
Personally I don't really know, If Google has ads, I just use gemini web for some vibe coding, I might take more questions to chinese models or even pay for them via something like open router where they might not have ads if this still happens via anonymity or just run the models locally yeah.
I wonder if we can create a community of people who have gpu who can have web of trust-like properties and they can share AI access to people who they trust and so on and maybe some gift properties but just niche closed properties where the trust factor is (more?)
Maybe offtopic but if we ever implement something like this, I really like https://ch.at 's interface and how minimalist it is.
It can't be done in general. "Hey Gemini, give me an exercise suggestion". When it suggests going for a run, how much of that is real and how much comes from the Nike campaign? Even google will not know the answer to where ads begin and end.
Advertisers might feel a bit weird about paying for something they cannot track, but for eg pharmaceutical companies have been taking doctors out to dinner for forever. Paying to sway the opinion of a communicator instead of paying to communicate is not necessarily new.
Depends on how deeply is the ad hidden. If the LLM-engine is using some deep level manipulations over a long period I'm not sure if blocking that ad would be possible.
What I forsee is a future of "local" AI engines. I think Apple is waiting for its moment for when people will ask for private-offline AI tools + a lot of smaller companies will follow. It won't be a huge success, but it will be popular in HN and in tech savvy crowd.
You presume that ads will just be near the content you want.. but what if ads will be served instead of the content, like instead of AI answering the question fully, some of it would be deffered to a book or instead of doing a trustworthy comparison it would bias the result?
People would run away, but in Gemini they're already drifting towards this direction with the useless garbage video suggestions..
