A visual map of the paper’s core claim: move from models that predict interfaces to models whose hidden state acts like the runtime itself. The page contrasts manifesto vs evidence, latent runtime vs external memory, and prototype strengths vs exactness failures.
Toggle between classical computing, memory-augmented neural systems, world models, and the paper’s proposed neural computer framing.
A compact radar-like SVG compares how strongly each paradigm collapses program, memory, and interface into one learned state.
From differentiable memory notebooks to world models to interface-conditioned rollouts. Hover nodes for details.
Heatmap of qualitative properties. The paper’s novelty is strongest in framing, weaker in demonstrated exact execution.
Switch between terminal and desktop prototype flows. Step through training signal, conditioning, latent state, and predicted rollout.
Mock evaluation grid showing where the prototypes look strongest: local visual continuity and action alignment, not durable symbolic semantics.
Bar chart of conceptual ambition versus what the transcript says the evidence actually supports.
Bubble plot of claims by ambition and evidentiary strength. Upper-right is the danger zone: strong rhetoric, thin proof.
Interactive failure map of the mismatch between approximate learned dynamics and exact symbolic computation.
Mock trajectories: local plausibility stays high for a while, but semantic fidelity decays faster as horizon increases.