Confused But Inspired

Projects

Silicon energy to bits

Nothing yet.

Markets conviction to price

Nothing yet.

Expression thought to form

Nothing yet.

Mind senses to model

  • cortexlive Oct 2026

    Predicts the human cortex’s fMRI response to any video, live, on a Mac. Three frozen backbones read the frames, sound and words; a ridge regression per participant maps them to 1000 cortical parcels, one volume every 1.49 seconds.

    Python · github.com/nihalssingh/cortexlive

    Four views of an inflated cortical surface. Each of 1000 parcels is shaded from blue to red by its predicted response, with low-reliability parcels in grey.
    Predicted response for one participant across a minute of a Friends episode, from 01:50.
    Two panels. Above, seven cortical networks as bands of red and blue across eleven minutes. Below, prediction accuracy for each network, from 0.200 for limbic to 0.399 for visual, each beside a lower speech-feature control.
    Seven cortical networks over the same episode, and accuracy by network against a speech-feature control.

Systems parts to behaviour

  • Interactive visualization of the causal light-cone effect. Watch a perturbation spread one hop at a time across a grid and trace the expanding cone of influence in space and time.

    JavaScript · runs in the browser

    Influence spreading from the centre of a grid.
  • Interactive visualization of spin-glass physics. Explore energy landscapes, flip spins, and discover ground states in planted Ising models.

    JavaScript · runs in the browser

    Spins flipped into the planted ground state.