You have a misunderstanding. Watermarking does not bias the responses in any way. How is this possible?

Before: "He leaped at the chance" - 33%. "Jumped at the opportunity" - 66%.

After: "He leaped at the chance" - 33%. "Jumped at the opportunity" - 66%.

But if you refresh your response from Anthropic 100 times:

Before: "Jumped at the opportunity" He leaped at the chance" "Jumped at the opportunity"

After: "He leaped at the chance" "He leaped at the chance" "He leaped at the chance"

The second one is detectable as being watermarked.

davmre has a good explanation that's more in-depth.

I understand, but this only applies at non-zero temperatures.

I've thought about this. Aristotle doesn't draw as hard of a line between right and wrong behavior or advocate for specific actions, with the right action depending on circumstance, and to finding the middle path. Traditional religions might be a bit more hardline, with specific rules or commandments. It's easier to identify with belonging to a particular group when the line between belonging and not belonging is more crisp and defined by particular practices.

Would love to know if others have takes.

These come from my own musings not someone else's - but bearing sycophancy in mind, I asked Claude and it told me this is basically the "strict church thesis" from sociology of religion, and linked me this paper, I think it lines up with my takes written from someone who has had this idea before: https://www.journals.uchicago.edu/doi/abs/10.1086/230409

It's more social, the translation of political is coming from polis; man is by nature meant to be part of a city or community.

I've advocated for this exactly. For alcohol it's called the three-tier system: https://en.wikipedia.org/wiki/Three-tier_system_%28alcohol_d...