Live data from Hacker News

LLMs can unmask pseudonymous users at scale with surprising accuracy

arstechnica.com

71–78 of 78 posts

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#71
post #56

There was a tool shared here that could show which accounts belong to the same person based on the writing patterns. Can't remember the name, but it found my old accounts on HN pretty accurately.

https://news.ycombinator.com/item?id=33755016 Way simpler than hnprofile from the sibling comment. This one used cosine similarity between user vocs - https://web.archive.org/web/20221126225241/https://stylometr...

Oh yes, this is the one I was referring to. Too bad it's shutdown.

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#72

Earlier quoted context omitted.

I'm pretty sure they can use the meta data the pull from your various interactions with search and the text you post online. These services build fingerprints of your habits using these techniques to follow you everywhere. At some point in the chain they could easily connect this fingerprint to your identity as soon as you log into and account that contains a piece of identifying information about you. The threat is…

Who is they? Which services? From what I can tell, the article/paper in question does not appear to utilize any of the techniques you mention, but I'd be interested to learn more about it. > it's the nuances behind the way you type I found this paper which talks about some of those methods. https://www.audiolabs-erlangen.de/content/04_fraunhofer/assi... For example the "Text" section on page 91.

[dead]

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#73

Earlier quoted context omitted.

The future is offline.

*selfhosted

I mean literally offline.

Computers were quite prevalent and useful before the notion of cloud computing, or the Internet.

Before the public was unleashed upon the Internet in 1992-1994, there were methods of information storage, indexing, searching, and exchange that didn't rely on real-time communications mediums operated by third parties. Example: CD-ROMs looked promising, the early 90's was smack in the middle of the "Multimedia" hey-day and gobs and gobs of data on nearly any subject was available and browseable at your perusal.

Of course it wasn't globally searchable, but there wasn't anything stopping anyone from making a master global index of CD-ROMs, selling it, and perodically updating it. Somebody (multiple somebodies) probably did. Libraries have been doing that for many decades. Replace chat with in-person meetups. Computer clubs were a thing in the 70s and 80s. DVDs still exist. DVD drives are $20 at my local Wal-Mart. SD cards are cheap and massive (1.5TB SD cards are a thing now).

Operating systems didn't always support TCP/IP. It's still something you can just turn off on a few of them.

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#74

Earlier quoted context omitted.

*selfhosted

I mean literally offline. Computers were quite prevalent and useful before the notion of cloud computing, or the Internet. Before the public was unleashed upon the Internet in 1992-1994, there were methods of information storage, indexing, searching, and exchange that didn't rely on real-time communications mediums operated by third parties. Example: CD-ROMs looked promising, the early 90's was smack in the middle of…

Pros and cons to this. It would re-centralize media again, stochastically dictating what we talk about. The Epstein case would have been sealed some five or so years ago without the onslaught of publicly visible interest in it.

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#75
post #52

Earlier quoted context omitted.

I’m curious if an LLM-based defense for this could be made. Like a browser plugin that warns you if you type identifiable information (like occupation) into a text field, and highlights turns of phrase that are “unusual” enough to be identifiable.

or something which just inserts random untrue details about you every now and again, like they do in Alaska, where I live.

Ah yes I remember seeing you at the Alaskan local underwater basket weavers meetup, you know the one for our profession.

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#76
Stylometry is just the most legible version of this. The harder-to-defend surface: posting time patterns, topic clusters, cross-platform phrase matching, interaction graphs. LLMs synthesize weak signals at scale in a way no single analyst could, which makes the threat model fundamentally larger than "change how you write." Most OPSEC advice is written for the pre-LLM world.

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#77
post #52

Earlier quoted context omitted.

I’m curious if an LLM-based defense for this could be made. Like a browser plugin that warns you if you type identifiable information (like occupation) into a text field, and highlights turns of phrase that are “unusual” enough to be identifiable.

or something which just inserts random untrue details about you every now and again, like they do in Alaska, where I live.

I'm afraid that if you do this, you won't just stand out among regular users, but you'll actually shine for such llm systems.

Re: LLMs can unmask pseudonymous users at scale with surprising accuracy

#78

Only if said users happen to commit OPSEC failures themselves. LLMs aren't magic... If someone can figure out who I am or what city I live in just by this username or my comments (with proof), I'll personally send you 500,000 JPY. I'm quite confident that's not going to happen though. The paper referenced in the article does not even explain their exact testing methodology (such as the tools or exact prompts used) be…

[dead]
Post reply on HN