The Naibbe cipher

[Pointless..]

A short reminder on the ai slop detection- they don’t, really.

According to the study below, all seven AI detectors unanimously identified 18 of the 91 TOEFL student essays (19%) as AI-generated and a remarkable 89 of the 91 TOEFL essays (97%) were flagged by at least one of the detectors.

You are not allowed to view links. Register or Login to view.

[nablator]

A short reminder on the ai slop detection- they don’t, really.

Some AI detectors were certainly bad in 2023 but some AI detectors are good in 2026.

[Koen G]

Like I said, when I don't fully follow a topic (calculation heavy) and someone using AI for translation posts about it, it becomes much harder for me to judge. Since Michael says the post makes sense, I don't think there's a problem.

[Pointless..]

A short reminder on the ai slop detection- they don’t, really.

If some AI detectors were bad in 2023 (true, I checked) it does not prove that all detectors are bad in 2026.

Right, it does not prove that the detectors in 2026 are worse or better.
But the study gives us reasons why the detection can fail. Those reasons seem to be still valid.

You are not allowed to view links. Register or Login to view.

But this is beside the point, in this thread. 
Back to the Naibbe cipher. An interesting approach in my opinion.

[ololololo]

At bottom, the big statistical limitation of the original Naibbe cipher is that each draw of a word (when encrypting standalone letters) or of a prefix/suffix (when encrypting letter pairs) is memoryless and independent.

I believe the reason for this is that Naibbe is based on the principle of substitution. A simple replacement of letters with vords...
I have a very poor understanding of your works. @magnesium, as I haven’t bothered to read them, so I’ll ask a somewhat strange question — what changes to the Naibbe cipher make it closer to VMS statistics? I.e., what needs to be done in an unintended way to make it even better (preferably without introducing rules based on the VMS text)?