Writings

  1. First page of Canvas Engineering: Declared Causal Macrostructure for Reverse-Diffusion Latent Dynamics
    ·research

    Canvas Engineering: Declared Causal Macrostructure for Reverse-Diffusion Latent Dynamics

    Prompt engineering structures what a model sees; canvas engineering structures what a diffusion model thinks in. You declare the latent regions, their connectivity, temporal frequencies, and loss roles as a typed schema, and a compiler lowers it into attention masks on a stock diffusion transformer — turning declared connectivity into an explicit causal graph inside the reverse-diffusion dynamics.

  2. First page of The Shape of Inquiry
    ·research

    The Shape of Inquiry

    A working draft on scientific agency as measurement construction: the hidden work of making the world answerable.

  3. First page of The Shape of Experience
    ·research

    The Shape of Experience

    A working book draft arguing that affect has geometric structure for biological and artificial systems navigating uncertainty.

  4. First page of Looped Attention in Video Diffusion Transformers
    ·research

    Looped Attention in Video Diffusion Transformers

    An empirical ROAM paper reporting 26 experiments and 236 training runs on looped attention, spatiotemporal canvases, CogVideoX-2B robot-video grafting, and the difference between parameter-efficient recurrence and actual iterative reasoning.

  5. First page of The Cortical Canvas
    ·research

    The Cortical Canvas

    A theoretical ROAM paper framing spatiotemporal canvas allocation as an architectural analogue of cortical real estate, local/global connectivity, temporal hierarchy, and adaptive recurrent refinement.

  6. First page of A Differentiable von Neumann Computer
    ·research

    A Differentiable von Neumann Computer

    A complete von Neumann computer rebuilt as differentiable tensor operations, shown by differential testing against three independent discrete implementations to be the same machine as its discrete counterpart, and demonstrated by running a self-hosting C compiler and a windowing OS. Reports what the construction established, what it cost, and why the synthesis question is still open.

  7. First page of Broadening and building beyond classical reinforcement learning
    ·research

    Broadening and building beyond classical reinforcement learning

    A survey-flavored note on what we lose by treating RL as scalar-reward maximization, and what alternatives a more honest framing suggests.