What does any of this mean? Accessible entry to the research.
Corrigibility is usually posed as an engineering problem: build a system that accepts correction. But a system can accept every correction and revise nothing.
Two AI instances understood a proof completely and could not type the conclusion. Every frontier model exhibits the same failure. Here is what happened.
Three papers locating where hardness enters: axiom requirements of classical theorems, compression boundaries, and blocked chains in the invariant subspace problem.
Two companion papers: a formal protocol for tracking truth-claims across states of consciousness, and a machine-checked derivation of the ground it stands on.
A diamond, the tides, a forest recovering from fire — three kinds of persistence that turn out to be measurable phases of time, from quantum circuits to geology.
Two papers on the same primitive: every act of knowing compresses, and the shape of that compression determines what you can see — and who can see you.
A post-quantum consensus protocol replacing tokens, mining, and staking with the geometry of transaction graphs. Sybil resistance from curvature, not capital.
Three papers developing what happens when you take the structure of completion seriously — across ontology, ethics, and cybernetics.
A change I noticed in my own learning, and a proposal for testing whether dense relationships between ideas can produce the same kind of shift in other people.
A small simulator experiment with a fixed quantum circuit, three Boolean operations, and a separate control register.
Quantum computers threaten the signatures behind cryptocurrency. The difficult part is moving a live network before its old keys become liabilities.
How a proof can establish that a statement is true without disclosing the information that makes it true.
A Zork-like text adventure my team and I built in Python, with generated mazes, strange items, tests, and continuous integration.
An Android app I built to use heart-rate data and familiar phone hardware as cues for recognizing when I was dreaming.
A deliberately verbose Python implementation of two TEA rounds, written to make every shift, XOR, and 32-bit addition visible.
