> Follow-up analysis of the first experimental results revealed a marked positional bias with LLMs tending to prefer the candidate appearing first in the prompt
Wow, this is unexpected. I remember reading another article about some similar research -- giving an LLM two options and asking it to choose the best one. In their tests LLM showed clear recency bias (i.e. on average the 2nd option was preferred over the 1st).
Wow, this is unexpected. I remember reading another article about some similar research -- giving an LLM two options and asking it to choose the best one. In their tests LLM showed clear recency bias (i.e. on average the 2nd option was preferred over the 1st).