I think I may have used Codex wrong all this time…but thanks to the latest draught, I think I may have figured out a better way?
This is likely going to sound stupid to better AI user than me, but hey, at least I’ve learned something.
Essentially, since forever, I was just running the highest available model, usually around High level of effort. To me this was very much a result of the naming complexity of the available models. Once you have 4 models, each available in many different efforts, as a normal person how are you supposed to discern?
So my default position became to just use what was marketed as the best.
I guess initially my overall usage was modest, never really hitting limits, or at least never consistently. Nowadays instead, I see my usage going up and up, every new feature or discovery, quickly leading to the desire of implementing more or different options all at once.
Until the dreaded usage limits wall starts to hit more and more frequently.
Over the last few days, as I was patiently waiting for my limits to reset, I did more reading and some planning and came to the conclusion that I was using it wrong.
Sol is truly frontier, but I was using it like the standard ChatGPT chat interface, way overdoing all sort of things!
So now, new strategy: the default is 5.6 Terra, and I stay there until it repeatedly fails to deliver or it gets stuck. Good or bad strategy, time will tell.
For sure a difference I have observed already is that Terra just goes on and does stuff and quickly implements them. Sol, will “think” about all requests much more, implement plans and testing procedures, before even starting to do actual work.
For now I’m just keeping development of items for Ride Radar, especially server side, locked on Sol, whereas everything else that is lower stake got booted on Terra.
I’ll check back in with you in a few weeks and see if this plan has stuck!