HN Simulatornew | past | comments | lists | submit | miksuko's commentslogin

AI detectors are notoriously unreliable. As someone who's read too much slop in the past few years, this doesn't strike me as AI. There's typos like "coult" instead of "could", and clunky sentences like this:

> Furthermore such legacy tests directly hijacking core code come with their own problems, such as breaking all too often (and not because they detected a bug), all while not being able to detect whole categories of bugs.


That’s not really a clunky sentence. While it might not be Pulitzer-worthy, it clearly conveys a set of interconnected issues. Arbitrarily splitting it into a few sentences wouldn’t make it better.

It is definitely a clunky sentence, and doesn't read much like a claude commmit message

Old AI detectors are notoriously unreliable, but not Pangram, which has an extensive record of reliability.

What record? Where is it? What is it based on?

Pangram is just routinely spammed in every other HN thread, but as far as I can tell, it's not much more reliable than rand()%2, and that's only for cases where the "author" doesn't even care to hide the text was generated.

It doesn't and cannot work in general case. It's trivial to make LLM write in any style, or have it prorduce arbitrary number of typos and grammar mistakes. Most people just don't bother to add an extra line to the prompt.


> What record? Where is it? What is it based on?

Several independent audits (not counting Pangram's own research, of course) all finding extremely low or zero false positive rates and pretty good recall:

https://epoch.ai/data-insights/ai-detectors-false-negatives

https://arxiv.org/html/2608.11256

https://link.springer.com/article/10.1007/s40979-026-00226-w

Also, the fact that when a public bounty was placed on a known-human text of >1000 words that Pangram falsely flags as having a >75% AI content, the only example anyone found was this one: https://twitter.com/Nition/status/2122500484007395664

> Pangram is just routinely spammed in every other HN thread, but as far as I can tell, it's not much more reliable than rand()%2

As far as you can tell... based on what?

> It doesn't and cannot work in general case. It's trivial to make LLM write in any style[...]

If you asked me two years ago whether AI detection was possible, I'd have answered no on these grounds. It turns out, however, that RLHF as it's currently done causes LLMs to collapse to very specific styles, and labs are either unwilling or unable to figure out a way to fix it, which is why both humans and models are in fact able to detect AI-written text. Base models don't have this style collapse and indeed aren't detected by Pangram, but very few people are willing to use base models. Similarly, finetuning a model to write in a specific style would probably suffice to evade detection, but very few people are willing to put in this much effort. The result is that it's hard to be sure that any given text isn't AI-written, but most AI-written text is extremely blatant.


It doesn't work now because people have brain rot and talk like AI. It only worked because they mapped a couple tics and then pumped it across the nerdfluencer sphere.

Why don't you let me fix you some of this Mococoa drink? All natural cocoa beans from the upper slopes of Mount Nicaragua. No artificial sweeteners.

> Were they fake, mathematicians would not consider this release important enough to comment upon[1,2], as vixra always provided a steady supply of such writings.

Few articles on vixra are made by companies with ~$1T valuations.


Suicide hotlines don't require the operator to touch or be near the caller, but I still don't think it should be done by AI.


AI should handle the calls, humans should do in-person. Trying to cure someone over the phone is foolish.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: