Documentation

Basics

AI Code Editing via a Tight, Symmetric Normalization Layer
An elaboration of the thesis, core arguments, and proposed implementation
Statements and Arguments: Foundation for "Intuition Based Prompting" and Context Bloat Prevention
1. Transformer-LLM Identity
On White Noise Prompting, Thalamocortical Analogues, and the Question of Machine State
An autoregressive language model does not know what it will say until it has said it. Each token is generated sequentially, committing the system to a path that constrains all subsequent tokens. There is no lookahead,…
Summary: LLM File Editing — Research State, Reasoning & Recommendations
The research field of Edit Representation for LLMs (also: LLM Code Editing) is nascent and rapidly evolving. It covers how changes to documents — not just source code — are best represented for language model generation.
1. How LLMs Process Your Prompt
Understanding what happens inside a language model when it reads and responds to your prompt is not just an academic exercise — it directly shapes which prompting strategies work, which fail, and why.
Text↔Code Symmetry in LLMs: Thesis, Core Arguments, and Proposed Test
The central claim is a statement about internal representation quality, not about output fluency:
Tokenizing — Statements and Arguments
Tokenizers convert text into sequences of numbers. Input text is broken down by a tokenizer into a chain of tokens — each token representing a word, sub-word, or character drawn from a fixed vocabulary. This numerical…

How-To Guides

RAG