> In all of these cases, the model's judgment is just very poor in a basic way.

This sentence resonated strongly with me. Not just in R&D but in every single interaction with the model. From Fable all the way to Sonnet they all very clearly show just zero judgement unless the answer was already seen during training.