I should publish a blog post titled “Using Antropic’s Claude is a Perversion of Writing”. If you are taking AI output and thinking it’s good and engaging, you are making a load-bearing mistake.

Joking aside, the criticism can come only after they implement it and we can see if it’s having any effect on the output quality. I doubt it’s going to be detectably worse.

If anybody is feeling so strongly against watermarking even before we can evaluate how it affects the output, I am only thinking that they have ulterior motives or they don’t want to be caught spreading copy-pasted slop.

Very interesting! Thanks for the links and the detailed answer!

Let me preface this by saying that I am generally against AI and I hate LLMs being pushed everywhere, but I currently hate facebook and social media even more than AI.

I think that ad blocking might actually be a good use of a model trained to detect ads: AI models are “black boxes” and it would make it difficult for facebook to find out precisely how the detection works and workaround it. Imagine a tiny classifier running locally, whose only job is to look at a post html or resulting rendered pixels and detecting if it’s an ad or not, and then generating the blocking rules.

It would be quite cool, because it would work on any website without an explicit list of ad-blocking rules that somebody needs to maintain.

Yes I never trusted it fully, and I would strongly suggest anyone to run it in a VM or at least a container, where it doesn’t have access to your credentials, browser cookies and sensitive files. It’s very irresponsible otherwise.

Example: around 1 year ago the bot made a small coding mistake and created a ‘~’ folder. I asked it to delete it (not in auto mode) and of course it went with a very nice “rm -rf ~” (which of course I denied).

Modern models are much smarter, but I would not trust the auto classifier to stop everything dangerous and make a similar hallucination slip through.