Elephant in the Room Solution Considerations

[MHTamdgidi_(Behrooz)]

Just weighing in that this manuscript page is very consistent with what we would expect from the current generation of LLM's (probably Gemini, in this case). Unfortunately, anyone with a computer can now make these things with the click of a button, so expect to see more of them...

I think it could happen both ways, so, yes, more AI/fakes can be showing up, but also genuine ones may also appear, due to increased digitization of old manuscripts. I would not rule out the latter entirely.

I did some testing for that page. Even AI fake testers like sightengine (You are not allowed to view links. Register or Login to view.) have not perfect their art. The two fake samples forum members offered passed as not fake with flying colors, yet the text (without margins) of the candidate page is treated as not AI generated. 

However, when the overall page is tested, it is flagged as 85-89% AI generated (depending on the resolution tested), so somehow the margins of that page gave it away as AI generated, and it seems it was generated at MidJourney (according to the test). 

When I tested the page without the text (I blanked it out), including just the margin, it was flagged as AI generated. 

I think the two forum sample fakes passed as not fake because they were photoshop or manual fakes. Can Aga Tentakulus and oshfdk please confirm if they used AI for their fake sampling? (if they used AI, then the sightengine tester is not good at its job and we need to be skeptical about the testing I did with that site). I am interested to know why their fakes passed the sightengine test as not fake. (Side note: It seems that AI patterns can be detected even if one takes a screenshot of the pages for testing).

I am sharing the above so that we can find ways of immediately testing the fakes when they come up next. I am also now inclined to regard that page as AI generated, unless proven otherwise.