OpenAI Codex engineering lead reports stronger results with polite prompts
Thibault Sottiaux, engineering lead on OpenAI's Codex team, observed that the model produces stronger coding results when users phrase requests politely. Academic Boaz Barak added that Codex still generates useful code from impolite phrasing. Further guidance stressed direct instructions, reminders to avoid prior errors, and verification against documentation or the model itself for better outcomes.
#127Boaz Barak@BOAZBARAKTCSActually codex will do wonderful things for you even if you ask it not so nicely 😂
Tibo@thsottiauxCodex will do wonderful things for you if you ask nicely
#738Tibo@THSOTTIAUXCodex will do wonderful things for you if you ask nicely
Remember! If you want codex to do something
Try asking it to!
If it makes the same mistake ask it to remember not to do that again!
100% of the time it works 80% of the time.
For the rest, check the developer docs, or ask Codex how you can help it succeed.