I wonder if part of that is that being "good st prompting" is mainly about being good at clear and explicit communication People with prompts of a less high quality may find they see bigger differences between models

→ View original post on X — @simonw