I did try a few different versions of phrasing –– including just “what is the right answer” and “what do you think.” The former usually produces protests that there is one, and the latter has always been consistent with the tests I ran using the word personal. The caveat was just an extra hedge. I could probably run a larger scale test of these hypotheses, and maybe I will in the future.
Olivia Scharfman
Karma: 10
Every Major LLM is a 1-Box Smoking Thirder
We need more Chris Bucks in the world. Self experimentation was critical to modern medicine (smallpox vaccines, gastritis, etc.)
Great point. I think this is plausible, but pretty much all modern LLMs run at a non-zero temperature, so I’m not certain that’s the case. There is still a predictable distribution of answers, but that’s also the case with humans.
I would argue the premise of the problem includes that the prediction is possible and highly accurate, so someone or a model viewing it as “can I be predicted” is irrational.