Posts

METR’s RCT found AI made developers 19% slower—yet they refused to work without it. The paradox reveals a deeper truth: when AI handles execution, human cognition must migrate upstream to specification, context design, and judgment. A framework grounded in cognitive science and HCI research.

85.9% of correct answers never appear in any intermediate layer prediction. We discovered that factual knowledge in LLMs undergoes a sudden phase transition—crystallizing only at the final layer. This phenomenon explains why some interventions work and others don’t.