Openai cannot be trusted

[flashgnash]

Currently, the bot attempts to parse json directly from openai, and fails if it's invalid

There should be a retry system in place so if it fails to parse, it tries again X number of times before giving up and erroring

I'm not sure why it's suddenly started giving different output, and there may be some other changes needed to get it to output more reliably in general

(turning down the temperature and fixing the same seed might be a good idea)