Yes. A little over a year ago I didn't trust LLMs to write code more than half a screen long. And sometimes it even produced nonsense there.
I still don't trust it for entire projects like the one yo...
Yeah, this one-off was an experiment. When I prompt Codex on a daily basis it is in very tight, very defined phases. My normal workflow is brainstorming with the best model of ChatGPT and developing a plan with phases, developing the definitions of each phase, and prompting Codex one phase at a time with tight definitions and restrictions.
This experiment makes me think I'm maybe overmanaging Codex a bit.
This experiment makes me think I'm maybe overmanaging Codex a bit.
1