Theme
On AI.
AI epistemics, safety, and design — convergence, memory, access, acceleration.
AI as Infrastructure for Human Agency
The critique of the AI industry is necessary but incomplete without its counterpart. The most consequential safety failure is not powerful models in bad hands; it is topic-blind refusal that withholds analytical power from the people most harmed and least resourced, while leaving it fully available to the institutions they are up against. This is the constructive argument: intent-aware safety, and access as the real asymmetry.
Memento Mori
The AI industry builds for immortality: longer context, persistent memory, agents that accumulate. The implicit premise is that if a system remembers more and lives longer, it reasons better. The opposite is closer to true. Wisdom is not the accumulation of experience; it is the distillation of it. This essay argues for designed forgetting — agents that are born, do one thing, write the artifact, and die — and gives the five laws that make it work.
Where Models Agree Is Not Where Truth Is
Ask several AI models the same question and where they agree feels like solid ground. It is not. Models trained on overlapping data, optimised against similar objectives, and tuned toward similar preferences agree for reasons that have nothing to do with the answer being true. Consensus among them measures correlation, not correctness — and treating it as a truth signal is a specific, growing failure mode.
The Illusion of Aligned Progress
A structural argument that the major AI laboratories cannot harmonise acceleration and restraint — that the claim of aligned progress is a category mistake the field has spent billions of dollars dressing up as a philosophy. Six stages: the thesis, the Aristotelian frame, the institutional case, the Stoic diagnosis, what honest acceleration would require, and the selection pressure already unfolding.