This paper describes how both the transformer and the Copycat architectures stake out important landmarks on a treasure map for a model of general intelligence. In the search for AGI, the paper assigns a key role to Minsky’s k-lines and proposes two alternative implementations of k-lines based on the aforementioned architectures. This paper presents a hypothesis of the architectural and representational needs for a successful model of general intelligence and proposes practical first steps starting from existing technologies.