Comment by wvenable
9 days ago
I wonder if it's time to step it up a little bit. We have all the LLMs that are great at looking at code and processing it so perhaps the future of ad blocking is letting the LLM handle it.
Has anyone thought about building something like this?
So you'd have a low-memory-usage super high latency ad blocker with a monthly bill, or an ad blocker that uses up several gigabytes of RAM/VRAM for slightly less high latency. Compelling.
I think GP is talking about using an LLM to write filters for the constantly changing Facebook ad serving code, which would be distributed to users.
Ah. I was envisioning the LLM handling requests as they come in lol
Yes, exactly this. The best way to fight AI is with your own AI, running locally on your own machine and loyal only to you. Also note that a highly-specialized ad-blocker agent wouldn't need a trillion parameters. It probably wouldn't even need a billion.
Hyperpersonalized code to its rescue. With the qwen3.8-27b release tomorrow and the advent of vibe coding and the improvements into proper agentic engineering we are moving towards hyperpersonalized software. This means we don't need a centralized maintainer for smaller tools. the facebook ad blocking cat and mouse game is something I cna very well imagine to just be something my computer would solve in real time, updating the uBlock filter/userscript to adapt to any fb changes in near real-time.