13 comments

[ 3.4 ms ] story [ 33.9 ms ] thread
> Boom, definitive "proof of typing" that a given piece was produced by genuine meat-on-keyboard effort.

Except it lacks proof of the keyboard - and the meat.

Yep, someone can "vibe code" something, that would emulate that human, the typing, the pauses, random typos, backspaces, and edits. There are probably models already available that describe the average delays between two consecutive keypresses depending on the location of keys, etc.
What if the human is just transcribing directly from AI generated text?
Trivial to simulate. Though I disagree that AI writing is impossible to differentiate from human prose, at least for now. It's still pretty obvious and still much worse than human prose (or, at least, much less interesting to read), though some are better than others, and I'm better able to spot writing from models I use regularly (Claude has a very distinctive style I can spot from a mile away, but that's true partly because I read its prose nearly every day when using it for coding).
A human should not have to or be compelled to prove to be human. The onus of proof here is the wrong way around, the premise fundamentally incorrect.
Didn't Pangram claim to have a >99% success rate?

https://news.ycombinator.com/item?id=48667761

From their homepage:

> Detect AI-generated content with 99.98% accuracy.

Are they wrong?

Though I ran the numbers, and even with a 0.02% false positive rate, that works out to about 6000 students falsely accused every semester, per university.

Lucid (https://www.writelucid.cc/) has a similar feature, though not for proving authorship, just for history. I don't know if you can definitively prove human authorship somehow.
I’ve been experimenting with a pet project that aims to solve this problem. “AI detectors” are certainly unreliable, I’m not sure they’ll ever get to a state where you can trust them.

I think concepts like this are the only reliable way to prove something was written by a human. A full replay like this is one way to do it. I think there are some other feasible ways to achieve this, maybe in combination with a full “replay”, but some sort of “proof of work” is the way to go I believe. As LLMs become more ubiquitous, I imagine products that solve the problem can be a real business opportunity.

Surveillance is worse than slop. Also seems like an disproportional use of resources compared to free formats and editors we already have today.

This product is in bad taste, and I hope it doesn't succeed.

AI writing is not writing and AI writers are also not writers.
people aren't going to read the trace, they're just going to rely on an AI grade to see if this was plausibly human written
I have been making recordings for some of the code I wrote since 2011, mostly because I like the visual effect of how the code came together, but I guess it also proved that I wrote those code manually. A recent example can be found here:

https://www.ioccc.org/2025/yang1/making.html