Hitherto : Notes from the Edge of Reason

Hitherto is a philosophical and ethical testing lab for language models. We probe not just what models know, but how they reason, why they fail, and where they gesture toward something like understanding.