4 ms·
was excited for a second thinking this was a way to use ai in real time to block elements on a page based on user defined criteria
by brigadier132 3y ago
was excited for a second thinking this was a way to use ai in real time to block elements on a page based on user defined criteria
- nerdjon 3y agoI was thinking the same thing when I read it. That honestly sounds like a fantastic use of AI that doesn't hurt anyone.
- baal80spam 3y agoHonest question - why something like that doesn't exist yet?
- Drakim 3y agoBecause detecting AI content is not yet a solved problem. And it might never be, because such a thing would be wonderful to train against while making the next big AI.
- adrianmonk 3y agoThe question directly above is about using AI as to enhance the content-blocking technology, not about using blocking technology to target AI content.
- minimaxir 3y agoThere's no performant way to do so, even with browser-friendly model optimizations.
- cadence- 3y agoIt probably exists, but not very well known, or is too expensive to use. For example, I wrote a SpamAssassin extension that adds some score to messages based on whether LLM thinks it is a spam email. It worsk very well, but it's a bit pricey when it comes to API costs. Maybe if I switch it from GPT3.5 to Claude Haiku it will make more sense. I think we still need to see prices of LLM tokens go significantly down and performance increase before such web filtering solutions make sense. The Claude Haiku is a big step int he right direction, but the price is still high for this kind of high-volume usage.