This is not a SFW instance. NSFW communities will be tagged, and efforts will be made to keep SFW communities clean, but proceed at your own comfort level.
LLMs are big statistical networks that determine the following word in a sentence. If two words are just as likely to follow, Claude is gonna pick the next word based on a secret key and somehow this can be looked up later. That appears to be the gist of it if I understood it correctly.
Yep, also my understanding of the mechanism. For small texts it’s probably not that hard to modify it enough to disguise it but the larger the text is, the harder it gets to remove the statistical watermark.
LLMs are big statistical networks that determine the following word in a sentence. If two words are just as likely to follow, Claude is gonna pick the next word based on a secret key and somehow this can be looked up later. That appears to be the gist of it if I understood it correctly.
Yep, also my understanding of the mechanism. For small texts it’s probably not that hard to modify it enough to disguise it but the larger the text is, the harder it gets to remove the statistical watermark.
It’s like a style, except it changes every word. If you know how, you can see when text is written in-style.