Please stop rewarding defensive code bloat in post training:
- try/except around code that can't throw
- null checks on values just created
- abstraction layers for one-off scripts
- "robust" looking verbosity over clean idiomatic code
Otherwise gpt5.5 was perfect
Show more
what would you most like to see improve in our next model?
