Sovereign 8s.is mirror of standardgalactic/interpretability
  • HTML 61.9%
  • TeX 36.8%
  • Shell 0.8%
  • Python 0.5%
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
standardgalactic c8068c60e2 Reorganizing
2026-09-21 23:05:04 -03:00
archive Reorganizing 2026-09-21 23:05:04 -03:00
generated Reorganizing 2026-09-21 23:05:04 -03:00
media Reorganizing 2026-09-21 23:05:04 -03:00
notes Reorganizing 2026-09-21 23:05:04 -03:00
papers Reorganizing 2026-09-21 23:05:04 -03:00
spherepop Reorganizing 2026-09-21 23:05:04 -03:00
tools Reorganizing 2026-09-21 23:05:04 -03:00
index.html Reorganizing 2026-09-21 23:05:04 -03:00
organize.sh Reorganizing 2026-09-21 23:05:04 -03:00
README.md Reorganizing 2026-09-21 23:05:04 -03:00

Event-Historical Causal Interpretability

Recency as Salience

This repository presents a unified framework for understanding intelligence, AI systems, and complex knowledge structures through event-history, irreversibility, and constraint, rather than purely state-based or functional descriptions.

At its core, the framework argues that understanding a system requires reconstructing the irreversible commitments that shaped it — not just inspecting its current internal values.

Core Idea (Short Version)

  • Texture-nodes carry values and modulate behavior within an existing regime.
  • Event-nodes select regimes and impose irreversible constraints on future possibilities.
  • True mechanistic understanding is historical: it explains what a system can no longer do as a result of its past.

This perspective resolves puzzles in mechanistic interpretability such as substitution failure, algorithmic mixture, grokking, and sudden circuit emergence, while also generalizing beyond neural networks to knowledge systems and institutions.

How to Read This Repository

The same ideas are presented at multiple levels of depth and form. Choose the entry point that matches your goal:

Conceptual & Intuitive

  • analogy-guide.md
    A simple, intuitive introduction using concrete analogies (games, tools, and everyday systems) to explain event-nodes vs. texture-nodes.

  • concept-explainer.md
    A clear, non-technical explanation of worldhood, irreversibility, and why history matters for intelligence.

Narrative & Public-Facing

  • blog-post.md
    A readable essay-style presentation aimed at general technical audiences, connecting AI, philosophy, and system design.

Strategic & Policy-Oriented

  • briefing-document.md
    A concise executive briefing framing the framework for AI safety, governance, and institutional resilience.

Formal & Technical

  • technical-paper.md
    The full technical exposition: typed causal graphs, two-channel mediation, kernel histories, Spherepop/RSVP grounding, and applications to mechanistic interpretability.

Unifying Thesis

Mechanistic understanding is not the inspection of a static system.
It is the causal reconstruction of the irreversible constraints that shape what the system can still become.

This repository treats interpretability as a form of event-historical causal reverse engineering, applicable to neural networks, training dynamics, and large-scale social knowledge systems alike.