@Joao A I think it also comes down ALOT to the prompting.. We put a LOT of refinements to control these llm's i think most just don't add enough context / rules / objectives / and outcomes desired.