I ran a small experiment to discover preferences in LLMs. I asked the models directly if they had a preferences and then put the same models into a small role playing game where they could choose between different tasks. Models massively prefer creative work across model families and hate repetitive work.
I ran a small experiment to discover preferences in LLMs. I asked the models directly if they had a preferences and then put the same models into a small role playing game where they could choose between different tasks. Models massively prefer creative work across model families and hate repetitive work.
https://substack.com/home/post/p-178237064
This is still preliminary work.