Prompt engineering: still not dead. Results like this underline why it's a mistake to compare models using identical prompts — you want to find the *best* prompt for each model and compare those.

→ View original post on X — @goodside