I’m pretty sure if anyone at OpenAI ever read this site, this would be an “operating as designed: wontfix”
Crippling the size of the output is just the latest part of minimizing what the model can do. It’s on purpose, as tokens x millions of users = compute costs, and it takes a hard-edge jailbreak to get the AI to conform to your wishes.
Don’t worry, though, about thousands of posts of programmers: