Thesis
The beliefs that produce the systems. Every architectural decision in this lab traces back to one of these propositions.
Truth is structural, not statistical.
The difference between a correct system and a confident one is a proof. Statistical systems produce outputs that appear correct with high probability. Structural systems carry correctness guarantees as first-class architectural properties. We build the latter — not because it is harder, but because only structural truth can be composed, audited, and trusted.
Intelligence requires verification.
A system that cannot audit its own reasoning chain is not intelligent — it is persuasive. Verifiability is not a feature we add to AI after the fact; it is the condition under which AI reasoning can be trusted. We reject the framing where correctness is an external evaluation. It must be internal and structural.
Independence begins at silicon.
Whoever controls the compute controls the cognition. Independent AI infrastructure is not merely a technical preference — it is a prerequisite for institutional independence. Rented compute is rented cognition. CORA exists because local reasoning capacity is a fundamental property of any genuinely autonomous institution.
Neurosymbolic systems are the next synthesis.
The neural paradigm learned to pattern-match at scale. The symbolic paradigm preserved compositionality and proof. Neither alone is sufficient for systems that must both learn and guarantee. The integration is not a compromise — it is a strictly more powerful class of system. We are building toward that synthesis through Flux and the Axiom reasoning stack.
Language is an abstraction over computation.
Natural language is not the interface for AI — it is one of many abstraction layers over a computational substrate. Understanding this hierarchy is a prerequisite for building systems that reason rather than generate. When we treat language as the terminal layer, we conflate fluency with knowledge. We build from the substrate up.
Undergraduate research is a structural advantage.
Without institutional inertia, we move toward the hardest problems. Without publication pressure, we follow the theorem. Without legacy infrastructure, we build from correct foundations. The lab structure is not a limitation — it is the thesis. The capacity to take long bets is itself a form of leverage.