Blogs
Attention Heads Learn Routing–Transport Orchestration
How trained attention heads pair routes with transports across model families.
Weight-Identifiable Antipodal GELU Populations Mediate Compensating Responses in Pythia
Finding compensating GELU circuits in Pythia from weights and testing their causal role.
Toward Algebraic Decompilation of Trained Transformers
A research program for recovering the computational architecture learned inside transformers.
DO NOT READ THE ANSWER
A novel about an AI, a sandbox, and the answer it was warned not to read.
Anthropomorphism and AI: When Human Intuition Clarifies, and When It Misleads
Which human concepts help us understand AI, and when do they lead us astray?
Protective Incompetence
How human limits quietly sustain trust, privacy, and the slack of civilization.
From nothing to something
Beginning with the empty set, a journey toward numbers, order, and arithmetic.