Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Large-Scale Online Deanonymization with LLMs

simonlermen.substack.com

Post reply on HN
Print view View original

101–110 of 258 posts

Re: Large-Scale Online Deanonymization with LLMs

#101

Post by monksy »

DrewADesign wrote: post #97
danielodievich wrote: post #88

I post under my real name here, pretty much the only place I post. It keeps me honest and straight in what I say when I choose to say it. I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don't know what it will be but I would expect some adversarial stuff. Trying to keep clean is what I'd prefer f…

> I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don’t think you’re wrong, but the fact that people consider it inevitable we’ll all have an immutable social acceptance grade that includes everything from teenage shitposts to things you said after a loved one died, or getting diagnosed with canc…

I think he's wrong and I'm willing to say that. The ability for people to move beyond the fundamental attribution error is well known and takes major resources to correct that. For anyone that posts a comment, assuming you want to have easy attribution later is that you must future proof your words. That is not possible and it is extremely suppressive to express yourself.

For example: "Ellen Page is fantastic in the Umbrella Academy TV show" Innocent, accurate, support, and positive in 2019.

Same comment read after 1 Dec 2020 (Transition coming out): Insensitive, demeaning, in accurate.

Re: Large-Scale Online Deanonymization with LLMs

#102

Post by KPGv2 »

danielodievich wrote: post #88

I post under my real name here, pretty much the only place I post. It keeps me honest and straight in what I say when I choose to say it. I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don't know what it will be but I would expect some adversarial stuff. Trying to keep clean is what I'd prefer f…

Fifteen years or so ago I read an article arguing that by the time Millennials are nearing retirement and have more political power, people will give less of a shit about what you did online in your twenties because we will have, out of necessity, learned that asshattery in your twenties is largely irrelevant to your trustworthiness in your sixties.

When I was that age, you could tell the kids who had political ambitions self-censored online. But now every is buck wild so you have to ignore that when looking at people.

For example, a MASSIVE portion of Millennials and younger looking at the Main election are pretty chill about the leading Democratic candidate having a Nazi tattoo because of this very thing. Basically, "dumb, drunk, deployed Marines will get cool skull and crossbones tattoos in their early twenties, and so what if he said a couple ill-worded somewhat misogynistic things in his twenties, that was decades ago, and he's obviously a different person."

Contrast with Bill Clinton, where he literally had to explain away university marijuana usage TWENTY YEARS AFTER THE FACT.

Point is, I think we're witnessing this evolution happening right now.

Re: Large-Scale Online Deanonymization with LLMs

#103

Post by bitwize »

Somebody I know irl has figured out I'm me here on Hackernews, based on the fact that my writing style here matches my verbal style. Fingerprinting people based on their words is one of the things I actually expect LLMs to be really absurdly good at.

Re: Large-Scale Online Deanonymization with LLMs

#104

Post by antonvs »

monksy wrote: post #101
DrewADesign wrote: post #97

Earlier quoted context omitted.

> I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don’t think you’re wrong, but the fact that people consider it inevitable we’ll all have an immutable social acceptance grade that includes everything from teenage shitposts to things you said after a loved one died, or getting diagnosed with canc…

I think he's wrong and I'm willing to say that. The ability for people to move beyond the fundamental attribution error is well known and takes major resources to correct that. For anyone that posts a comment, assuming you want to have easy attribution later is that you must future proof your words. That is not possible and it is extremely suppressive to express yourself. For example: "Ellen Page is fantastic in the…

> Same comment read after 1 Dec 2020 (Transition coming out): Insensitive, demeaning, in accurate.

I genuinely don't understand this. Are you sure you're not imagining possible offenses against some non-existent standard?

Re: Large-Scale Online Deanonymization with LLMs

#105

Post by slopinthebag »

danielodievich wrote: post #88

I post under my real name here, pretty much the only place I post. It keeps me honest and straight in what I say when I choose to say it. I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don't know what it will be but I would expect some adversarial stuff. Trying to keep clean is what I'd prefer f…

I think as the younger generations come of age they simply will not care about that sort of thing. Like it or not, it's part of the culture and might just be accepted as the norm.

Re: Large-Scale Online Deanonymization with LLMs

#106

Post by pbhjpbhj »

ectospheno wrote: post #96
danielodievich wrote: post #88

I post under my real name here, pretty much the only place I post. It keeps me honest and straight in what I say when I choose to say it. I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don't know what it will be but I would expect some adversarial stuff. Trying to keep clean is what I'd prefer f…

I expect more people over time to use local LLMs to write every single post they make online.

>post they make

Will they realise their life has devolved to pretending an LLM is them and watching whilst the LLM interfaces {I was going to say 'interacts', not this fits!} with other bots.

Will they then go outside whilst 'their' bot "owns the libs" or whatever?

Hopefully at some point there is a Damascus road awakening.

Re: Large-Scale Online Deanonymization with LLMs

#107

Post by observationist »

danielodievich wrote: post #88

I post under my real name here, pretty much the only place I post. It keeps me honest and straight in what I say when I choose to say it. I tried talking to my children about leaving as clean of a footprint on the internet as one can in anticipation of future people/systems taking that into consideration. I don't know what it will be but I would expect some adversarial stuff. Trying to keep clean is what I'd prefer f…

Autonomous Proxies for Execration - spam bots whose entire purpose is flooding the internet with spam so as to make identifying anything true utterly impossible. If you can't differentiate between real and unreal information in online comments, then online comments stop being a significant factor in shaping public opinion. You need to abstract - identify reliable sources of information, individuals or institutions that do the work to collect and curate.

We're already seeing this as a side effect of the mishmash of influence operations on social media - with so many competing interests, mixed in with real trolls, outrage farmers, grifters, and the like, you literally cannot tell without extensive reputation vetting whether or not a source is legitimate. Even then, any suggestion that an account might be hacked or compromised, like a significant sudden deviation in style or tone or subject matter, you have to balance everything against a solid model of what's actually behind probably 80% or more of the "user" posts online.

There are a lot of aligned interests causing APEs to manifest - they're a mix of psyop style influence campaigns, some aimed at demoralization, others at outrage engagement, others at smears and astroturfing and even doing product placement and subtle advertisement. The net effect is chaos, so they might as well be APEs.

Re: Large-Scale Online Deanonymization with LLMs

#109

Post by ranger_danger »

famouswaffles wrote: post #35
ranger_danger wrote: post #23

Earlier quoted context omitted.

But you also relied on people giving away too much personal information about themselves... which won't always be the case.

Over a large enough timeframe (often a couple years at most), almost everyone online gives too much information about themselves. A seemingly innocuous statement can pin you to an exact city and so on.

I would be quite impressed if someone could figure out what city I live in from my 4.5 year old account, but I highly doubt it.

Re: Large-Scale Online Deanonymization with LLMs

#110

Post by ranger_danger »

DalasNoin wrote: post #46
ranger_danger wrote: post #23

Earlier quoted context omitted.

But you also relied on people giving away too much personal information about themselves... which won't always be the case.

I agree that these accounts probably on average still contain more information than the average pseudonymous account. I think we could try to use the LLM to increasingly ablate more information and see how it performance decays – to be clear we already heavily remove such information, see Table 2 appendix. But I don't expect that to change the basic conclusions.

I also wonder how well the LLM would do with less direction e.g. just ask it to analyze someone's posts and "figure out what city they live in based on everything you know about how to identify someone from online posts".
Post reply on HN