Rome/Writing
Featured Writing
An empty writing desk covered in manuscripts, lit by a single hanging lamp — Obra Dinn 1-bit dithered.
2026Jun 10

Everyone Who Can Appreciate a Symphony

On becoming a judge in the age of AI — and why that might be, in some domains, the same thing as becoming a creator.

ai philosophy creativity epistemology
Read piece →
AI Scientists Are Here
2026 Mar 06

AI Scientists Are Here

Claude Code + Tmux + Modal ran nine hours of autonomous ML research. The results are real.

ai research agents
8 min Read →
№ Filed Title & abstract Read
  1. 05 2026Jul 07

    Twelve Thousand Games in Avalon

    Twelve thousand games of The Resistance: Avalon say the mission board is the least informative surface in the game. What finds a liar is the talk—and Evil's most disciplined habit turns out to be a mistake it pays ten points for.

    games data deception interpretability philosophy-of-mind
    Read →
  2. 04 2026Jul 07

    Your Agent Config Is Infrastructure

    A practical map of an agent setup — canonical config, skills worth building first, subagent delegation with git worktrees, a second model, a few plugins worth adding, a notes vault, and a feedback loop that writes corrections back into the skills — concrete enough to copy.

    agents tooling workflow
    Read →
  3. 03 2026Jun 28

    What the Router Learns

    GORGO is a cross-region LLM request router that scores cache locality, replica load, and wide-area network latency in one cost function—and learns the weights from the deployment’s own latency stream.

    llm-serving systems routing performance
    9 minRead →
  4. 02 2026Jun 28

    Absolute Light: From Ratios to Nits

    Why HDR so often looks wrong — and what the file is really telling your screen. Relative vs. absolute light, the PQ transfer function, and the Rec.2020 gamut, with drag-to-compare HDR demos.

    imaging hdr display signal-encoding
    13 minRead →
  5. 01 2026Jun 04

    Tracing Thought in Language Models

    Mechanistic interpretability tools that reveal how Claude's internal circuits execute reasoning, plan outputs, and implement safety mechanisms.

    writeup interpretability mech-interp SAEs attribution-graphs
    Read ↗
  6. 00 2025Feb 15

    The Ecstasy of the Proof

    On comprehending a mathematical proof as an ecstatic experience — a transfiguration from ambivalence and skepticism to conviction.

    philosophy mathematics phenomenology
    22 minRead →
Follow

Follow on X →