When the Canonical Completion Is Wrong: Formalizing and Measuring the Jump in Large Language Models
This paper attempts to develop a formal account of the jump in four steps and measures the second, and hopes it paves the path for formalizing and measuring the full jump in the future.