[pulkas]

article is too old. who is using gpt-4o today?

[_0ffh]

That's a valid concern, given the paper makes clear that the effect over the polite/impolite scale seems to be model dependent (it finds the reverse correlation of earlier studies on even older models).