Centralised curation will have its time but marketers will move to systems which are unique to each visitor. Think 30 second ads which are unique to each visitor, built on their shopping history, social media comments, search, etc. These systems are currently prohibitively expensive to build and deploy, but they won't be with cheap LLMs. YouTube is already experimenting with ads inside streams, for example. Those are very difficult or impossible to deterministically block. Sites will get very good at agentically evading deterministic adblocking software. This is where local LLMs will shine. Consider the case of YouTube in-steam ads. The LLM could probabilistically determine that the current frames are ads (even though this particular ad is 100% unique to that visitor, and has never appeared in any ad database before), then black out the section of the screen/audio while it plays. Or fast forward until it finds where the segment ends. All in real-time, locally determined.

It does become a strange cat and mouse game in the future, but I feel quite strongly that ad blocking will win. Users will have intelligent, autonomous systems on their devices which can curate and block content as desired.

It burns through tokens/electricity.

In all seriousness, far too many people think "throw AI at it" will solve every tech problem. AI remains inefficient, power hungry and mercurial. Replacing an adblocker or firewall with a constantly-running AI is as efficient and useful as rocket motors on a car.