[carefulfungi]

How close are we to a third party with sufficient compute being able to mass-de-anonymize social media? What happens if they republish social media feeds with identity probabilities? Do we reach a point where the internet is anonymous to casual users but not to large corporations or governments? Presumably someone is already selling identity-labelling as a service?

Amusingly, (locally generated) LLM text becomes an anonymity mask in those scenarios.

[kevin_thibedeau]

I use real name here and posted about pending litigation once. An insurance settlement was offered a week later. Socials are scraped to build profiles for these scenarios. People engaged in fraud tend to blab about it and many business are interested in having such evidence accessible. It can be useful to exploit that system to your advantage when the truth is on your side.

[robotswantdata]

See http://arxiv.org/abs/2602.16800

And http://news.ycombinator.com/item?id=33755016

Also http://github.com/antirez/hnstyle

Glhf

[b00ty4breakfast]

I was thinking about this the other day, using LLM text to thwart stylometric analysis. At some point, I'd read about using machine translations to do the same thing (translating a text to, for example. Chinese then back to English). This seems to only work if the translation method isn't quite perfect so get an "Engrish" effect or the like. But you could probably feed your manifesto or whatever to one of the various modern chatbots and have it rewrite it in the style of Poe or Pynchon or maybe a generic business email. (Obviously, setting aside the issue of if the chatbot is keeping all this stuff in a database somewhere).

[negura]

i've wondered the same recently and came across this resource:

https://www.whonix.org/wiki/Surfing_Posting_Blogging#Stylome...

tl;dr intentionally writing in a different style is very effective and overall the best option

btw one upside to public content being ingested in LLM training data. is that certain writing styles (like using the emdash) become much more common, due to reappearing in generated content. this means that if you continue to write in that style, you will blend in more easily. which makes your own content resist stylometric analysis

[threepts]

Just as close we were 10 years ago.

CIA has always had this privelege.