This study comes about the heels of people routinely complaining that GPT-4 has subjectively declined in performance over the past few months. Popular theories about why include OpenAI "distilling" models to cut back their computational overhead inside of a quest to speed up the output