Historically, LLMs have struggled with long contexts and certain output types, and have tended to favor lengthy outputs.

“We have seen historically they have issues with long contacts and they have issues with certain kind of outputs. They prefer, let's say, lengthy outputs and stuff.”