Skip to content
View ameritusweb's full-sized avatar

Block or report ameritusweb

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ameritusweb/README.md

Hi, I'm Scott

I build AI architectures from first principles — often in C#, usually from scratch — to test ideas that don't fit the "scale it and see" paradigm. Then I write up what I learn at Scott's Thoughts, my Substack on AI architecture, interpretability, and grounded cognition.

The through-line across everything here: I think the interesting open problems in AI are structural, not scale-bound — how intelligence is grounded, how it compartmentalizes, how we can read what it's doing from the physics of the computation rather than the geometry of its activations.

The code and the writing go together

Most of my repos are working experiments behind an essay:

Repo What it is The essay
ArtificialBiologicalIntelligence Growing intelligence from endorphin/pain receptors up to grounded language via evolutionary receptor topology — 138 receptors, 16 families, 46 steps Growing Intelligence From Scratch
thermodynamic-epistemic-routing 98.4% factual/speculative classification with zero inference-time parameters, by measuring forward-pass energy The Thermodynamic Interpretability Breakthrough That Scales
ParallelReverseAutoDiff PRAD — parallel reverse-mode automatic differentiation in C# for building custom neural networks. The engine under most of my ML work Why Scaling Laws Break Down
Meritocious A discussion platform where AI curates for substance instead of karma
prologscript An AI reasoning framework built on causal-diff semantics Cause and Effect is a Diff, not a Chain

I also build other things for fun — cloudcad (web-based CAD with Babylon.js + CadQuery), games, transpilers, UI frameworks.

Outconceive

Outconceive is my studio — where the bigger builds live:

  • 🎲 outconceive.com — unconventional strategy games. First up: Warp 1 Chess, dimensional chess on overlapping boards, with an AI trained on 174,000 self-play matches that discovered strategies I didn't design in.
  • 🧵 pst-os — an operating system built on Parallel String Theory, in Rust. The essay behind it: Position is Identity in OS Design.
  • 🏥 followup — a mobile app for whoever accompanies a patient to the ER: it interviews you while you wait and produces a timestamped, plain-language observation record for the triage nurse.

Where to find me

📝 Substack: ameritusweb.substack.com — essays on AI architecture, structural safety, and where scaling breaks down. If any repo here made you curious, the writing is where the ideas connect.

🎮 Outconceive: outconceive.com — where the inconceivable becomes playable.

Pinned Loading

  1. cloudcad cloudcad Public

    Web-Based CAD with Babylon.JS and CadQuery

    JavaScript 9 4

  2. ParallelReverseAutoDiff ParallelReverseAutoDiff Public

    Parallel Reverse Mode Automatic Differentiation in C# for Custom Neural Network Development

    C# 9 1

  3. Meritocious Meritocious Public

    Meritocious is a new kind of discussion platform — a space for thoughtful, high-quality conversation curated by AI, not karma.

    C# 2

  4. ArtificialBiologicalIntelligence ArtificialBiologicalIntelligence Public

    Growing intelligence from endorphin/pain receptors to grounded language through evolutionary receptor topology selection. 166 receptors, 20 families, 60 steps. The organism builds the world that bu…

    Python 2

  5. thermodynamic-epistemic-routing thermodynamic-epistemic-routing Public

    98.4% factual/speculative classification with zero inference-time parameters — by measuring forward-pass energy, not activation geometry.

    Python 1