this post was submitted on 06 Mar 2026
16 points (75.0% liked)

Privacy

5416 readers
141 users here now

Welcome! This is a community for all those who are interested in protecting their privacy.

Rules

PS: Don't be a smartass and try to game the system, we'll know if you're breaking the rules when we see it!

  1. Be civil and no prejudice
  2. Don't promote big-tech software
  3. No apathy and defeatism for privacy (i.e. "They already have my data, why bother?")
  4. No reposting of news that was already posted
  5. No crypto, blockchain, NFTs
  6. No Xitter links (if absolutely necessary, use xcancel)

Related communities:

Some of these are only vaguely related, but great communities.

founded 1 year ago
MODERATORS
otter@lemmy.ca
shaytan@lemmy.dbzer0.com
16
LLMs can unmask anonymous internet users for $1–4 each, matching 67% of pseudonymous Hacker News accounts to real LinkedIn profiles at 90% precision (arstechnica.com)
submitted 2 weeks ago by Innerworld@lemmy.world to c/privacy@lemmy.dbzer0.com
22 comments fedilink hide all child comments
top 22 comments
sorted by: hot top controversial new old
[–] lvxferre@mander.xyz 9 points 2 weeks ago (1 children)

If it was just BullshitIn and Brainrot News I'd laugh my arse off. The problem is the rest of the internet.

And it gets worse than what the article says. It isn't just the information present in the text; it's also how you use language. Spelling, word choice, punctuation, average sentence size, even discourse structure, those are all things that can be used to associate 2+ accounts as coming from the same person, they're like a linguistic fingerprint. And neural networks are rather good at finding this sort of pattern.

Fuck, you don't even need a NN to do that. It's possible to do it by hand, just incredibly boring and laborious.

[–] Corkyskog@sh.itjust.works 6 points 2 weeks ago (2 children)

That's why I fractured my personality into 6 seperate entities!

[–] lvxferre@mander.xyz 3 points 2 weeks ago (3 children)

That's one way to do it, I guess. For most people it's simpler: develop a second writing style and only use it online when sharing sensitive information — such as when whistleblowing or sharing pirate content. That writing style should:

  1. Prioritise brevity. The less you say, the harder it is to tie your online personae.
  2. Stick to the standard spelling of the language you're using. Typically there's only one standard way to write something, but multiple non-standard ways.
  3. Avoid as much as reasonably possible words and grammar that reveal your location, social class, education level, age, or any neurodivergence you might have.
  4. Only use three types of punctuation marks: comma, period, interrogation.
  5. Keep text structure simple.

Small note regarding people who speak 2+ languages: keep in mind each language you know interferes on your usage of the other languages, often in predictable ways. Even your native one. And this can be used to narrow down a bunch of profiles to who you are.

[–] Techlos@lemmy.dbzer0.com 6 points 2 weeks ago (1 children)

Ironically, one of the most effective methods is to use a local LLM to rewrite your text with a low temperature setting, moving your writing style to the mean of the LLM style. Erase your identity by hiding as a bot.

[–] Corkyskog@sh.itjust.works 2 points 2 weeks ago (1 children)

Wait... Unless your using a local LLM aren't you basically giving all the info your trying to hide to the spying bots?

[–] moosh@slrpnk.net 2 points 2 weeks ago

yeah it must be local for security and privacy

[–] naevaTheRat@lemmy.dbzer0.com 1 points 2 weeks ago

Humans are not generally capable of such separation. If you do want to do this sort of thing then you should probably use a tool. Of course your tool will likely mark you, but at least it wont lead back to you without the tool. Like the equivalent of blowing out security cameras with a bright LED.

For instance you could use a script that identified all words with synonyms that are used less frequently and replace them with the most common, that removes all spelling errors, that corrects grammatical error, and highlights sentences that run on for editing.

Generally if you want to avoid ID invent a fake person and being that person is easier. Also like, don't share things, don't comment, and ablate all character from your writing (amusingly slop bots are fantastic for removing all humanity from your writing, but general tone analysis tools can suffice).

[–] hector@lemmy.today 0 points 2 weeks ago

Interrogation as punctuation? Idk what that is. Commas, periods, semicolons are underused, I have no idea what the former is.

[–] hector@lemmy.today 1 points 2 weeks ago

Other me is jealous of that many instances. That guy is a dick though.

[–] solrize@lemmy.ml 6 points 2 weeks ago (2 children)

Stay off LinkedIn.

[–] mnemonicmonkeys@sh.itjust.works 4 points 2 weeks ago

Good thing I've never made a post on LinkedIn. The only thing I've done is search for jobs on there

[–] naevaTheRat@lemmy.dbzer0.com 2 points 2 weeks ago (1 children)

attacks like this are just correlation/looking for distinct high info signals. Lemmy is if anything even more vulnerable via federation. But you knew that, because you read the article and didn't just reply like a 5 year old struggling against the irresistible call of a marshmellow.

[–] autonomoususer@lemmy.world 1 points 2 weeks ago* (last edited 2 weeks ago) (1 children)

When's the last time you put your real name, current location, employer, coworkers and work history on Lemmy?

[–] naevaTheRat@lemmy.dbzer0.com 0 points 2 weeks ago (1 children)

I didn't say that.

Lemmy, and all federated socmed, is much easier to data mine. The way you talk, the spelling mistakes you make, when you post etc can all be used to locate you and demographically identify you and lemmy replicates everything, even posts you delete or edit, to anyone federated.

[–] autonomoususer@lemmy.world 1 points 2 weeks ago* (last edited 2 weeks ago) (1 children)

Getting on LinkedIn don't fix that.

[–] naevaTheRat@lemmy.dbzer0.com -1 points 2 weeks ago (2 children)

Do you talk to people like this in real life?

[–] ParlimentOfDoom@piefed.zip 1 points 1 week ago

Nice try, AI. Not gonna identify me that way

[–] autonomoususer@lemmy.world 1 points 2 weeks ago

Cut the personal attacks, like yours on the original commenter, and people in real life will talk to you too.

[–] naevaTheRat@lemmy.dbzer0.com 4 points 2 weeks ago

It's mentioned in the article but we live in an age of anti intellectualism and apparently nobody reads. People have been able to do this with high accuracy for a long time, like at least 10 years.

This just requires less expertise.

[–] kittykillinit@lemy.lol 2 points 2 weeks ago (1 children)

Are there any ways to do this for free?

[–] naevaTheRat@lemmy.dbzer0.com 2 points 2 weeks ago

With a basic knowledge of statistics, some time, and caffeine, all things are possible.

If you mean script kiddie it I fucking hope nobody has shipped that lol

[–] Astertheprince@lemmy.dbzer0.com -1 points 2 weeks ago

This is a very misleading headline, this isn't really possible with people who use good OPSEC, and yes that means sharing identifiable information on LinkedIn. Also the developers aren't showing their work claiming it will be misused. Which leads me to believe this is a combination of people having really bad OPSEC and also fearmongering.