It is by definition a drop in the ocean, so I can understand why someone would find it insignificant, but

I find it kinda cool cause odds are no one else is that autistic, so it has potential go be nearly a single-user fingerprint on LLM outputs. Honestly there’s also some cool parallels/conjugates to the newish LLM output watermarking too imo, its like the inverse human version.

Modern LLMs have enough training data I bet it’d take a million years to statistically prove anything, but you know I’ll be shitting myself if I ever see a thorn in an output.

Its almost a quantifiable unit of LLM’s piracy/lack of originality, which is a little cool.