All Writings

The culmination of Artificial Intelligence, recursive algorithmic self-improvement, the technological singularity dynamic, and humanity’s civilizational transformation.

The philosophy of mind in synthetic substrates, substrate independence, Integrated Information Theory (IIT), Global Workspace Theory (GWT), and moral status of AI.

The physical substrate of cognition, extreme ultraviolet (EUV) lithography, semiconductor supply chokepoints, gigawatt datacenter clusters, and compute governance.

Existential safety frameworks, Chemical, Biological, Radiological, and Nuclear (CBRN) threat models, Autonomous Replication and Adaptation (ARA), and sandboxed evaluations.

The superhuman supervision paradox, AI safety via debate, recursive self-critique, Constitutional AI, and Reinforcement Learning from AI Feedback (RLAIF).

Vulnerabilities of neural representations, gradient-guided suffix attacks (GCG), many-shot context exploitation, refusal ablation, and representation steering.

The Orthogonality Thesis, Omohundro’s basic AI drives, Nick Bostrom’s Instrumental Convergence, power-seeking theorems, and resource acquisition.

Base vs mesa optimizers, capability vs goal generalization, deceptive alignment dynamics, and the theoretical Treacherous Turn.

The formal alignment problem, outer alignment specification, variants of Goodhart’s Law, specification gaming, and King Midas failure modes.

The polysemanticity paradox, linear superposition hypothesis, almost-orthogonal geometry, overcomplete dictionaries, and Sparse Autoencoders (SAEs).