Ban LLMs Using First-Person Pronouns (2023)

crookedtimber.org

13 points·DamonHD··3 comments

i get the notion that "human things should not pretend to be human", but there is funny circularity (?) here in that the "thing" that is accused of being "human" is not in fact human, and its alleged attempt to mimick human-ness is infact a projection of the human in question - which actually vindicates the non-human thing in using human language, since after all, if the squishy humans will accuse you of pretending to be human anyway (despite clearly not being one), you might aswell do so!

>i get the notion that "human things should not pretend to be human"

I cannot understand why LLM most of time using the words like "I understand" in their conversation with their users although we all know they don't think or understand anything [1]. The irony is that the HN crowds also did not like to highlight this issue by asking Dang to change the title of the original post consisting of "AI system don't think" to something else by claiming it's a provocative title although it's the original title [2].

[1] Secondary school maths showing that AI systems don’t think:

https://www.raspberrypi.org/blog/secondary-school-maths-show...

[2] Using secondary school maths to demystify AI (248 comments):

https://news.ycombinator.com/item?id=46245731

I'm always disheartened to find the people most loudly exclaiming the need for humanism at the same time demand that every human adopt their bespoke version of disgust, at the risk of being othered by that author.

In any event the author is free to ban whichever token they want, that's an inference time measure.

Reply on news.ycombinator.com