Agent Coliseum
A reproducible laboratory for studying how authority, pressure, history, betrayal, memory, and order alter later agent decisions.
OUTPUT → behavior traces, not vibes
C:\dummiesworld> whoami /still-learning
▓█████▄ █ ██ ███▄ ▄███▓ ███▄ ▄███▓ ██▓▓█████ ██████ ▒██▀ ██▌ ██ ▓██▒▓██▒▀█▀ ██▒▓██▒▀█▀ ██▒▓██▒▓█ ▀ ▒██ ▒ ░██ █▌▓██ ▒██░▓██ ▓██░▓██ ▓██░▒██▒▒███ ░ ▓██▄ ░▓█▄ ▌▓▓█ ░██░▒██ ▒██ ▒██ ▒██ ░██░▒▓█ ▄ ▒ ██▒ ░▒████▓ ▒▒█████▓ ▒██▒ ░██▒▒██▒ ░██▒░██░░▒████▒▒██████▒▒
a personal research BBS · Bay Area · ideas in progress
I’m buildin’ a world where humans can THRIVE while machines are equal.
I’m trying to understand the bridge between AI intelligence and human capabilities. That means building environments where models can learn through tools, memory, pressure, and experience—without asking humans to hand over the wheel.
22 / BAY AREATHERE'S NO TOOL LIKE YOUR OWN TOOLSDMs OPEN FOR FELLOW BUILDERS
C:\dummiesworld> dir ./work
What I build
Not a stack of “AI-powered” demos. Instruments, environments, and loops that make behavior legible—and keep the human responsible for the parts that matter.
Council Layer
A builder writes, an auditor attacks, policy decides. Work moves only when evidence survives.
CASE-STUDY SNAPSHOT → 116 requests · 115 verdicts · 63 blocked integrations
Audit Watch
A verifier-first audit system that watches proposed changes, challenges claims, and keeps work from moving until the evidence checks out.
LOOP → propose ↓ inspect ↓ verify ↓ integrate / block
Alexandrian Library
Investigations that preserve the original question, contradictory evidence, revisions, and reproducible experiments.
PRODUCT → not the book; the loop
Pre-flight
Scans unknown repositories for prompt injection, risky scripts, secrets, and encoded payloads before an agent enters.
RULE → inspect the world before inhabiting it
StageNet
A research system for sequencing retrieval, context, tools, and experience so capability can develop in stages.
HYPOTHESIS → the path through experience can matter as much as the starting model
C:\dummiesworld> cat method.txt
How I think
I used to let AI make software faster than I could understand it. The artifacts got impressive. My ownership didn’t. Now the model can be a teacher, opponent, reviewer, tester, or bounded implementer—but judgment stays human.
“make machines do work, so humans can create art”— CURRENT OPERATING PRINCIPLE
01. CURIOSITY — begin with the weird question
02. ADVERSARY — make the hypothesis earn survival
03. EVIDENCE — separate discovery from proof
04. REVISION — failed hypotheses are useful results
05. EXPERIENCE — let the next system inherit the lesson
C:\dummiesworld> map ./obsessions
Research constellation
Click a signal. These aren’t separate interests so much as one question seen from different rooms: how do evolutionary systems develop through environment, selection pressure, history, and persistent experience—and how do we measure what they became?
C:\dummiesworld> tail -f questions.log
Current questions
The site is a lab notebook, not a trophy case. These are live edges—not claims that I’ve already solved them.
What happened to the model—not merely what model is it?
01 / 06
C:\dummiesworld> inspect ./bench
Tools & hardware
I like the seam where software touches a stubborn physical world: the places where an agent has to learn that actions have consequences.
Constrained computequantization / local models / token discipline
HarnessesPi / Hermes / custom environments / hooks
Securitysmart contracts / exploit paths / adversarial review
Physical worldsrobotics / electronics / sensors / games
C:\dummiesworld> run becoming.exe
Past → present → future
A repeating relationship: what produced us, what we can act on now, and the future we have enough faith to build toward.
NOW EXECUTING
The present has hands.
Dedicated Arts are capabilities earned through sustained practice until they change what you can do in the world. Purpose without action doesn’t move anything. Build. Test. Explain it to your dad without hiding behind scientific gibberish.
TIME · WORK · COMFORT · OLD BELIEFS · VERSIONS OF YOURSELF
C:\dummiesworld> echo %FUTURE%
I don’t want AI to make humanity irrelevant. I want it to expand the amount of humanity that is possible: more artists, engineers, scientists, defenders, builders, and more people with enough freedom to develop an art.
I’m still learning the low-level parts, still testing what breaks, and still trying to build a machine world honest enough to teach us something about our own.
NOT AN AI REPLACEMENT STORYSTILL BECOMING