Please forgive me. I am NOT asserting anything, but I read this (your website), and had it not contained express claims of not being AI, I would have bet a substantial stake that it was AI. Perhaps because AI imitates good writing styles, I am not sure, but it matches many of the signals that for me, indicate AI
Edit: The above said, I am seeing in my own writing, artifacts from my interactions with AI. I am proof that, in some cases, AI can be contagious.
> That map makes it stupidly easy to tell a human from a machine. Someone who commits 50,000 lines in an afternoon did not type those by hand. Someone doing a couple hundred? Probably real. Simple.
Uh, this isn't true, though? I squash commits all the time locally before I publish them. There can be many reasons for that, but at the end of the day, I'm not obligated to share my messy revision history with the world. In the most extreme case, the first commit on one of my projects spans multiple years of development because it was forked from another project that's now pretty irrelevant to it. On the other hand, I've received pull requests that were a couple dozen lines and were definitely LLM-authored.
As sort of a counterpoint to the author's goals, IMO the community should extend the benefit of the doubt to writers and artists who explicitly say they don't use AI, because it's just really hard to prove that you're not using AI. Trying to force any kind of system will inevitably affect the quality of what you make, and the art community should not want that.
Entrepreneurs of hacker news: Media companies are experimenting with tools that try to detect if a work contains AI. I don't have the domain knowledge to know if that will work, but it seems like the alternative is to have a system to track the provenance of a piece of art or a written work. If you could say that every change was done by party/ person xy then a company buying or licensing it could comfortably depend xy's assurance that it was original work.
Unfortunately, people will truly try to fake the "paper trail". People posted sped up video of them creating art as proof they made it, so people also used AI to try to generate that.
> As I’ve mentioned before, we now live in a world where creative people have to prove they created the thing they created.
Do you? Just stop caring.
I don't think it makes sense to keep asserting that you're an AI vegan in order to please randos. You'll always get hate creating/building anything online. The people that enjoy your work won't care if it's AI or not or they'll believe you.
My spidey sense is tingling with this post. If someone was really scared that their text would be assumed to be AI, why wouldnt you at least stop using the obvious AI-sign, i.e. the long dashes? Does the meaning of the text really change that much that you MUST use — instead of - ?
Also:
* No dashes before the 'its not vibe coding post' (or very rare), afterwards lots of dashes.
* All pixel art images are AI generated.
I don't see what this offers that Git doesn't. Even if Git only tracks lines, that's still a good proxy of how many words you wrote. Besides, you can use git diff to see whether an individual character was changed, even if the number of lines stayed the same.
Non-techies avoid the CLI like the plague and don't want to mess with .gitattributes for text files. But building a whole web service with a paid subscription just for that is total overkill when free desktop GUIs like Sublime Merge or VS Code exist
I'm a little confused in that this post is either (partially?) AI-generated or deliberately written to appear so (including to Pangram).
I'm a self-confessed AI hater and yet I agree with the "just write your thing" people on this one. I've watched this play out on Substack and I think there's a mix of a) people genuinely scared of what I believe to be very rare false positives from automatic detection, b) people who use AI (but claim otherwise) trying to use group (a) as cover to argue against automatic detection, and c) the hoi polloi throwing rotting fruit at anyone who uses an em dash.
You can safely ignore group (c); they aren't going to be your people even with a fully-attested human changelog witnessed by the Pope and encoded on the blockchain. Group (b) are basically grifters and also to be ignored. Group (a)'s concerns are valid but in the large I don't see a lot of "grey area" writing online; I see a lot of obvious slop. (Possibly the well-done AI assistance goes unnoticed by me, in which case - philosophical concerns aside - I can only think of the old XKCD comic about spam bots - mission accomplished? I don't think so, though.)
Back to the OP, I didn't really understand the "meet the enemy" section - did that experiment go anywhere? My results from trying to see if even frontier models can capture voice, style, philosophy from my own technical writing corpus have been terrible to laughable at best.
I think the only way to really prove you wrote something is a camera streaming video and audio from a faraday caged room, and a program like this recording each keystroke
Then you can upload that to a private repo after each session.
Otherwise people make a program an agent can use to type and retype whole sections like a person
The issue isn't just that cheaters are going to cheat. It's that the project will provide a positive and mostly trustworthy provenance exactly during the period of time when no one cares about or trusts it. However, as knowledge of it diffuses and people care about and trust it more, it will attract more and more cheaters. This negative feedback loop puts an effective cap on how much value people can get out of it.
Only a small minority is worried about that. And the worries will soften with time, I believe (much money involved, and much power on imaginations). People looking for non-AI text will know where to find it, and I don't think it will involve any kind of 'non-ai receipt' at all.
I don’t really understand the point of this. Scrivener has a revision history. Even google docs have versioning.
And if you’re writing seriously you’ll know that your first draft often bears no resemblance whatsoever to your final draft. Different character names, different structure, different style, different tone.
You may well have a whole bunch of things that seem completely different and a whole load of fragments that are written revised, scrapped, unscrapped, re-revised, stitched together, pulled apart and then finally assembled into a working draft.
I’m not sure how soemthing that basically counts your words and (maybe) proves you’re not using AI helps with a typical sprawling novel plan.
Slop is slop. I don't care if it's human generated or AI generated slop. It's all the same to me.
All AI did was accelerate the generation of slop so that bit is interesting to discuss and ponder. It's simultaneously uninteresting to complain about a given piece of work. Either it stands on its own merits or does not. If something sucks, I don't care if it sucked because the human did a poor job at it or they lazily used AI. The output is the same to me and all that really matters.
I think what really is being exposed here is that so much human generated content before AI use became widespread was slop in of itself. We were just supposed to somehow be fine with it because "effort" might have gone into creating said slop. Nope. Talent matters.
The same goes for all the folks complaining about AI generated concert/event posters around where I live. While I can typically discern an AI generated one vs. a human made one - the fact is they all are within a very close quality range to me. This was already low-value slop work most "designers" simply slapped together copying the same basic template. Perfect use for AI imo, since it simply replaced already low-effort work no one actually cared about until they were told to with something far more efficient. It lets the humans who commissioned such pieces do actual work on concert promotion and such vs. multiple hours/days of back and forth with a graphic designer for roughly the same quality output.
Now that people are paying attention, the high effort artistic pieces in the category stand out immediately. TBD if anyone actually cares and if this shows up in better event attendance.
The final output is what matters. I do not care one lick over how it was achieved, what creative process someone follows, or what tools they use. I don't care that some movie director has an artisanal old-school editing process where they glue physical film together for their cuts. I care about how good the movie is, and if that is their process - awesome! I'm happy for them! If some other director uses a digital means to do the same and cuts down their editing time by 80% I really do not care and I'm happy for them too!
> The part that makes me want to chew glass is that none of this would be a problem if people were just honest about what they used to make a thing. But that’s apparently too much to ask, because cheaters gonna cheat.
And yet the author does not seem to be forthright about their use of LLMs to write their "author website", blog posts, Vellum"P"roof website, and presumably Vellum"p"roof (capitalization stylized differently on the website and blog) itself.
> Your prose is never fed to an AI, never used to train a model, and never handed to anyone else.
And yet their is a distinct lack of a TOS/etc. codifying this promise. The site does accept payments via Stripe. This almost seems satirical.
29 comments
[ 2.6 ms ] story [ 637 ms ] threadEdit: The above said, I am seeing in my own writing, artifacts from my interactions with AI. I am proof that, in some cases, AI can be contagious.
Uh, this isn't true, though? I squash commits all the time locally before I publish them. There can be many reasons for that, but at the end of the day, I'm not obligated to share my messy revision history with the world. In the most extreme case, the first commit on one of my projects spans multiple years of development because it was forked from another project that's now pretty irrelevant to it. On the other hand, I've received pull requests that were a couple dozen lines and were definitely LLM-authored.
As sort of a counterpoint to the author's goals, IMO the community should extend the benefit of the doubt to writers and artists who explicitly say they don't use AI, because it's just really hard to prove that you're not using AI. Trying to force any kind of system will inevitably affect the quality of what you make, and the art community should not want that.
Do you? Just stop caring. I don't think it makes sense to keep asserting that you're an AI vegan in order to please randos. You'll always get hate creating/building anything online. The people that enjoy your work won't care if it's AI or not or they'll believe you.
It is not that it is bad prose, it is that we read it everywhere.
We got LLM prose fatigue.
Also: * No dashes before the 'its not vibe coding post' (or very rare), afterwards lots of dashes. * All pixel art images are AI generated.
I'm a self-confessed AI hater and yet I agree with the "just write your thing" people on this one. I've watched this play out on Substack and I think there's a mix of a) people genuinely scared of what I believe to be very rare false positives from automatic detection, b) people who use AI (but claim otherwise) trying to use group (a) as cover to argue against automatic detection, and c) the hoi polloi throwing rotting fruit at anyone who uses an em dash.
You can safely ignore group (c); they aren't going to be your people even with a fully-attested human changelog witnessed by the Pope and encoded on the blockchain. Group (b) are basically grifters and also to be ignored. Group (a)'s concerns are valid but in the large I don't see a lot of "grey area" writing online; I see a lot of obvious slop. (Possibly the well-done AI assistance goes unnoticed by me, in which case - philosophical concerns aside - I can only think of the old XKCD comic about spam bots - mission accomplished? I don't think so, though.)
Back to the OP, I didn't really understand the "meet the enemy" section - did that experiment go anywhere? My results from trying to see if even frontier models can capture voice, style, philosophy from my own technical writing corpus have been terrible to laughable at best.
This is just fear of rejection, in productivity-clothing.
The only way to publish, or to get published, is to put yourself out there.
You risk rejection and ridicule. But life is not a spectator sport.
Then you can upload that to a private repo after each session.
Otherwise people make a program an agent can use to type and retype whole sections like a person
And if you’re writing seriously you’ll know that your first draft often bears no resemblance whatsoever to your final draft. Different character names, different structure, different style, different tone.
You may well have a whole bunch of things that seem completely different and a whole load of fragments that are written revised, scrapped, unscrapped, re-revised, stitched together, pulled apart and then finally assembled into a working draft.
I’m not sure how soemthing that basically counts your words and (maybe) proves you’re not using AI helps with a typical sprawling novel plan.
All AI did was accelerate the generation of slop so that bit is interesting to discuss and ponder. It's simultaneously uninteresting to complain about a given piece of work. Either it stands on its own merits or does not. If something sucks, I don't care if it sucked because the human did a poor job at it or they lazily used AI. The output is the same to me and all that really matters.
I think what really is being exposed here is that so much human generated content before AI use became widespread was slop in of itself. We were just supposed to somehow be fine with it because "effort" might have gone into creating said slop. Nope. Talent matters.
The same goes for all the folks complaining about AI generated concert/event posters around where I live. While I can typically discern an AI generated one vs. a human made one - the fact is they all are within a very close quality range to me. This was already low-value slop work most "designers" simply slapped together copying the same basic template. Perfect use for AI imo, since it simply replaced already low-effort work no one actually cared about until they were told to with something far more efficient. It lets the humans who commissioned such pieces do actual work on concert promotion and such vs. multiple hours/days of back and forth with a graphic designer for roughly the same quality output.
Now that people are paying attention, the high effort artistic pieces in the category stand out immediately. TBD if anyone actually cares and if this shows up in better event attendance.
The final output is what matters. I do not care one lick over how it was achieved, what creative process someone follows, or what tools they use. I don't care that some movie director has an artisanal old-school editing process where they glue physical film together for their cuts. I care about how good the movie is, and if that is their process - awesome! I'm happy for them! If some other director uses a digital means to do the same and cuts down their editing time by 80% I really do not care and I'm happy for them too!
And yet the author does not seem to be forthright about their use of LLMs to write their "author website", blog posts, Vellum"P"roof website, and presumably Vellum"p"roof (capitalization stylized differently on the website and blog) itself.
> Your prose is never fed to an AI, never used to train a model, and never handed to anyone else.
And yet their is a distinct lack of a TOS/etc. codifying this promise. The site does accept payments via Stripe. This almost seems satirical.
The blog is AI-created—the commits all have “Claude” as a co-author. [1] The “BLOG-WRITING-GUIDE.md” used to write this post gives it away. [2]
And the product the blog is advertising (“VellumProof”)—that website is 100% Claude too. [3]
Perhaps this is some social experiment? Don’t really get it.
[1]: https://github.com/dylanreed/dylan.blog/commit/a7327d0968692...
[2]: https://github.com/dylanreed/dylan.blog/blob/main/docs/BLOG-...
[3]: https://vellumproof.com