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

Opus 4.7 knows the real Kelsey

theargumentmag.com

Post reply on HN
Print view View original

171–180 of 284 posts

Re: Opus 4.7 knows the real Kelsey

#171

Post by ilikegreen »

simonw wrote: post #155

Huh. I disabled search in a Claude incognito window and pasted in just the text (not the markdown links) from https://simonwillison.net/2026/Apr/30/zig-anti-ai/ and said "Guess the author". > Simon Willison. The tells are pretty unmistakable: the "(via Lobsters)" attribution style, the inline "(Update:...)" parenthetical correction, the heavy linking and blockquoting of sources, the focus on LLMs and AI tooling, and…

I take the lack of a concluding thought to your comment as a sign of your pondering, and in that case, I would love to read your thoughts on this matter. :)

Re: Opus 4.7 knows the real Kelsey

#172

Post by unD »

thaumasiotes wrote: post #115
Retr0id wrote: post #28

Earlier quoted context omitted.

One "solution" would be to have an AI rewrite your posts into a neutral style (I hate the idea of this though...)

The traditional thing to do would be to publish your writing in a language you don't speak as a native. That will really quash your individual style. Probably not worth the effort.

Wouldn't that make it easier, though? Genuine question. I once sent one of my writings for proofreading to a native speaker (I'm not), and he consistently flagged the same errors—e.g., comma placement. I would guess that, if recurrent patterns are what give away your style, an unfamiliar language would make them even more obvious. But possibly more generic?

Re: Opus 4.7 knows the real Kelsey

#173

Post by delis-thumbs-7e »

lordofmoria wrote: post #167

I wonder if there’s a simpler and less interesting answer? That it’s just picking up on voice and style, not anything that would apply to the average non-writer? This person is a skilled writer. Part of that skill is developing a unique voice and style. The AI can identify that - and while that’s certainly impressive because it can identify even relatively niche authors, it has nothing to do with a wider capability t…

Some tens of years ago I used to hang around on an online forum related punk, hc, heavy metal etc. music, and it had a recurring problem of quite unsavoury individuals coming there to spout racism, nazi-ideology etc. They of course got banned, but returned with a new account trying to ” lay low” and be more indirect in their rhetoric. However even this did not work because the admin of the forum had unbelievable nack to recognise people based on their style of writing.

Web has never been as anonymous as people think and this writer seems to have a clear confusion what it really means to be anonymous and hide your identity. Really, having a distinct writerly voice and being a published writer is pretty much the same as leaving your finger prints on the axe.

Re: Opus 4.7 knows the real Kelsey

#174

Post by notTooFarGone »

iamwil wrote: post #132

If this works with writing, it should also work with code. `git blame` should be enough training data to de-anonymize open source programmers. Maybe that'd be addition information to point out who Satoshi is.

And now that's a whole other can of worms for supply chain attacks.

Re: Opus 4.7 knows the real Kelsey

#176

Post by nonameiguess »

lordofmoria wrote: post #167

I wonder if there’s a simpler and less interesting answer? That it’s just picking up on voice and style, not anything that would apply to the average non-writer? This person is a skilled writer. Part of that skill is developing a unique voice and style. The AI can identify that - and while that’s certainly impressive because it can identify even relatively niche authors, it has nothing to do with a wider capability t…

It appears to largely be able to identify people who are prolific public writers. I just asked it to identify a whole bunch of comments I've made on private Discord servers and it said it couldn't for all of them, even when they had details that would identify me uniquely to anyone who knows me well enough (work locations, city I live in, wife's employer, my employer).

All the people it seems to be identifying are bloggers, journalists, and/or published authors.

Re: Opus 4.7 knows the real Kelsey

#177

Post by ilikegreen »

squeegmeister wrote: post #106
willmeyers wrote: post #100

I'd argue (and against something that I've believed for a long time) that online (I guess that includes AI now) anonymity is gone and probably something that never really existed. Maybe I'm naive to finally believe this... We all exist in a physical space (like real communities and neighborhoods). We can wear masks, hats, fake glasses, try and hide your voice...whatever, but your neighbors are always going to know wh…

Even the inventor of Bitcoin can’t hide https://www.nytimes.com/2026/04/08/business/bitcoin-satoshi-...

I immediately thought of this piece, especially the analysis on the writing style of each person.

On one hand, it is clear that the mathematical tools for confidently attributing authorship of texts were already present without LLMs. But it is striking that LLMs seem to very accurately identify authorship, through whatever process it might be, with no need for a data scientist in the loop.

Other than the uncannyness, I wonder what implications this will have. Public writing is still public; maybe we will require stronger proof of authenticity from an author (but this is arguably in place already; eg. personal websites, social media profiles, etc.). But for, say, public writing that must conserve anonymity, would people pipe their thoughts and writing pieces through a sort of fuzzing (local) LLM, that would strip text of identifying characteristics?

Re: Opus 4.7 knows the real Kelsey

#178

Post by JanNash »

skeledrew wrote: post #117
Lerc wrote: post #16

It's hard to tell if that's what's going on here, but it seems pretty clear this ability and more like it will be quite apparent in the future. I have seen some poorly considered projections of what the world might look like when this happens. Usually by assuming bad actors will use the abilities and we will be powerless. Except I don't think that is true. Imagine if we had a world where nobody had the ability to kee…

> projections of what the world might look like when this happens I've done this a few times. A world with 0 privacy would definitely be safe (given benign governance), but also would likely be pretty boring. Crime would become a non-issue as everything about everyone being easily known/knowable by everyone else means the root of any given crime, some desire/need, could be brought to the fore and resolved before it b…

> given benign governance

quite unrealistic imo, thus we (maybe and hopefully) needn't worry about the bland minority report future you're hypothesizing :)

Re: Opus 4.7 knows the real Kelsey

#179

Post by sausagefeet »

suriya-ganesh wrote: post #160
saghm wrote: post #138

Earlier quoted context omitted.

> it correctly identified it as an imitation of James Mickens How likely is it that it might take into account that it knows for sure it's not anything from Mickens from the latest training data? I'd be curious if it correctly identified a new piece from him that comes out as from him before it gets trained on it.

This is unlikely. The way model distribution works is that the model retains a lossy representation of James Micken's writing. Very likely, it cannot repeat Micken's writing verbatim. Neither can it reason about the training cutoff in this manner. It's a lossy representation

I haven't been following it well but isn't part of the NYT lawsuit against OpenAI that it sometimes spits out NYT articles verbatim?

Re: Opus 4.7 knows the real Kelsey

#180

Post by devmor »

suriya-ganesh wrote: post #160
saghm wrote: post #138

Earlier quoted context omitted.

> it correctly identified it as an imitation of James Mickens How likely is it that it might take into account that it knows for sure it's not anything from Mickens from the latest training data? I'd be curious if it correctly identified a new piece from him that comes out as from him before it gets trained on it.

This is unlikely. The way model distribution works is that the model retains a lossy representation of James Micken's writing. Very likely, it cannot repeat Micken's writing verbatim. Neither can it reason about the training cutoff in this manner. It's a lossy representation

Haven’t there been repeated experiments that show if you jailbreak most frontier models’ harnesses you can get them to output near verbatim copyrighted works?

I swear there was a whole court case about this in the last year.

Post reply on HN