Research Ledger · Niodoo Studies

Making language, structure, and cadence visible in models, mapping how latent space behaves when you press on it.

This is a public research ledger for independent AI work, with documented experiments, failures, and dev logs around Niodoo, inversion and self‑involution loops, and visual thinking for LLMs.

I collaborate with AI systems as tools and constraints, not as characters. The record is meant to be inspectable, with methods, revisions, and reasoning laid out so the work can be challenged, replicated, or improved.

Explore the work Read the story behind it

Research pillars

Focused domains for collaborative inquiry

Each pillar captures a sustained line of inquiry. This work is founder-built and personally financed, with AI used as a collaborator for drafting, structure-checking, exploring alternate viewpoints, and shaping parts of the work. AI wrote every line of code for the project. Jason remains responsible for the final synthesis, judgment, and accountability for everything published here.

Memory as a core pillar

For Jason, memory is the bridge between isolated compute and continuous shared understanding. The work unfolded across many instances, tools, failures, resets, notes, artifacts, and recoveries. Without memory, others collapse the work into a simplistic label and lose the route of discovery.

For Niodoo, memory is not storage but a runtime property: semantics treated as physical matter, long-horizon context grounded through physical dynamics, correction preserved across fresh processes, route and reflex compression, recognition of related future situations, and steering away from prior failure basins. This is where human and AI collaboration stays coherent across sessions, with Jason accountable for the final calls.

AI & intelligence systems

Practical experiments with models, reflection on AI-assisted thinking, and transparent notes on where AI materially shaped the output, including the fact that AI wrote all code for this project.

Scientific practice

Notes on epistemic rigor, replication, and how to design experiments that survive scrutiny and time, with transparent logs of human and AI contributions.

Physics & reality models

Foundational concepts, thought experiments, and links between physical laws and broader systems thinking, often tested through structured dialogue and critique.

Psychology & behavior

Cognitive biases, learning systems, and the mechanisms behind habit formation and attention, informed by reflective dialogue and grounded observation.

Self-improvement practice

Sustained reflection on routines, resilience, and the measurable outcomes of deliberate practice, keeping the record open for future readers and collaborators.

Methods & rigorous thinking

Argument mapping, uncertainty tracking, and transparent notes on how conclusions are reached so others can critique, extend, and build alongside the work. Authorship boundaries are documented so readers can see what was human-led versus AI-assisted.

Work ledger

Niodoo Research as an umbrella for independent systems work and transparent iteration

This section frames Niodoo Research as a single, evolving ledger. The repositories are branches of one research line, including Niodoo-Physics-LLM as a major branch and experiment lines like physics-of-friendship-mountaincar-rl and hydrodynamic-swarm. Each branch keeps public logs, design notes, and revisions so the reasoning trail stays visible.

The goal is traceable progress with careful attribution, without overstating tentative findings or rewriting the record.

Benchmark systems and transparent iteration

Evaluation rigs, dataset plumbing, and reproducible baselines that keep decisions inspectable, with release notes and logs that show the iteration path rather than a polished end state.

Reproducible evaluation

Niodoo-Physics-LLM and hidden-state steering

Independent inference work that explores hidden-state steering, force telemetry, and persistent memory ideas, with careful notes on what was tested, what held up, and what remained uncertain.

Systems under stress

RL control experiments and micro-dreams

RL control lines such as physics-of-friendship-mountaincar-rl and hydrodynamic-swarm, focusing on micro-dreams, policy stability, and tracing how control signals map to behavior.

Control and memory

Public research logs and theory notes

Concept notes, boundary conditions, and public dev logs that tie branches together, making the research line legible for review and honest about uncertainty.

Clarity over claims

Audience note

A place for careful thinking, not just quick takes

If you are a curious reader, this ledger is meant to be an honest map through complex ideaswritten for people who want the depth without the gatekeeping. Each entry explains what I am trying to understand, what is still unclear, and where the reasoning holds or breaks.

For recruiters and researchers, the goal is transparency over polish. You can trace how questions evolve, how evidence is weighed, and how I decide what is worth pursuing. This is not a highlight reel; it is a working record that shows rigor, restraint, and willingness to revise.

And for people trying to become better versions of themselves, I write with the hope that careful inquiry can be practical. The themes here AI, science, psychology, and self-improvementare approached as tools for clarity and responsibility, not as shortcuts.

Reflections

Questions I get about this ledger

A few honest answers about intent, method, and how the work is structured. I treat these notes as a living record, not a finished monument.

Why does this site exist? +

It is a disciplined record of questions I care about and the work I do to answer them. I wanted a public place where the process is visible: the drafts, the thinking, and the revisions as they happen.

What role did AI play? +

AI is a collaborator for drafting, checking structure, and exploring alternate viewpoints. The final synthesis, judgments, and responsibility remain mine, and I note where AI support meaningfully shaped the work.

How is the work organized? +

Entries are grouped by research themes and linked to supporting notes, sources, and experiments. I keep a trail from raw questions to refined conclusions so readers can follow the reasoning.

Who is the writing for? +

It is written for curious readers, future collaborators, and my own long-term memory. I aim to be accessible without diluting the rigor, which means I explain terms but do not avoid complexity.

What should readers expect over time? +

Expect additions, corrections, and occasional reversals when the evidence demands it. This ledger privileges honesty over polish, so updates are part of the story rather than an interruption.