Glossary

Terms from the learning system: encoding, retrieval, and WPW.

Agent Memory

Durable context that an LLM agent can read and update across sessions.

Active Learning

Learning where the learner transforms information through comparison, evaluation, explanation, mapping, retrieval, or application.

Compiled Wiki

The synthesized knowledge layer produced from raw sources and prior outputs.

Bear Hunter System

The user’s main encoding system. It turns new material into a structured map through inquiry, investigation, and refinement.

Deep Processing

Mental operations that build meaning by connecting information into durable, usable schemas.

Dimensions of Learning

The central study-system model: Deep Processing, Self-Management, Self-Regulation, Mindset, and Retrieval.

Kolbs Experiential Cycle

A reflection cycle for turning experience into better experiments through reflection, abstraction, and action.

Marginal Gains

A method for choosing small, compounding improvements that contribute to a larger goal.

Health Check

A maintenance pass that finds gaps, contradictions, stale pages, orphan pages, and unsupported claims.

Metacognition

Awareness and control of one’s own thinking processes during learning or problem solving.

Mindset

How a learner interprets difficulty, feedback, mistakes, identity, and growth.

Retrieval

The active reconstruction and use of knowledge before checking a source.

Self-Management

The systems for time, tasks, focus, energy, habits, and environment that make it possible to carry out learning.

Self-Regulation

The control process of monitoring learning, diagnosing gaps, and adjusting strategy.

Spaced Interleaved Retrieval

A retrieval system that mixes recall, increasing delays, varied contexts, and gap repair.

WPW

A whole-part-whole retrieval method for testing both big-picture structure and detailed recall.

Passive Learning

Learning behavior where the learner consumes or records information without meaningfully transforming it.

Raw Source

Original or near-original material used as evidence, stored under raw/.

Source Note

A metadata-rich markdown file describing a source, its claims, relevance, and compilation status.

Working Memory

The limited mental capacity used to hold and manipulate information during thinking, encoding, and retrieval.

Global Workspace

A limited-capacity buffer that broadcasts selected information so that it is available across the whole system. It is the leading cognitive-science model of conscious access, and it is closely tied to working memory.

Access Consciousness

Information being globally available for reasoning, report, and control. This is the functional sense of “conscious,” distinct from subjective feel (phenomenal consciousness).

J-space

In Anthropic’s interpretability work, the small set of internal patterns in a language model that carry its deliberate reasoning. Each of these patterns stands for a word the model is about to say. The J-space is the model-side analogue of a global workspace.

J-lens

Anthropic’s tool for reading the J-space. For every word in the model’s vocabulary, it finds the internal pattern that makes the model more likely to say that word later, averaged across many contexts.

Functional Component

A Chinese character part classified by its function in one specific character: form, meaning, sound, or empty. The same shape can have different functions in different characters.

Form Component

A character component that contributes by depicting something. It does not indicate a sound or a sense.

Meaning Component

A character component that contributes to the character containing it an abstract extension of the component’s concrete original sense.

Sound Component

A character component that indicates the pronunciation of the character containing it as a syllable range. It rarely gives the exact reading and almost never gives the tone.

Empty Component

A character component that indicates no sound and no meaning in a given character. It is usually a distinguishing mark or a remnant of historical corruption.

Sound Series

The set of characters sharing one sound component. These characters are related through seven groups of initials, and the initials in each group share a place of articulation.

Semantic Series

The set of characters sharing one meaning component. It is used to work out the range of senses that component has across those characters.

Meaning Tree

A structure for a character entry. It derives every sense from the original meaning, marks loan senses as off-tree, and gives each reading its own tree.

Original Meaning

The sense a character’s form was created to write. Later senses can be derived from it, so they do not need to be memorized.

Fence Principle

When an authoritative statement says that two readings are related, a learner accepts and absorbs sound connections that the learner would otherwise reject without noticing.

Rebus Principle

Borrowing an existing character to write an unrelated word that sounds the same. It explains many apparent component mismatches. The story of the borrowing can also be used as the mnemonic.

Pipelining (Character Intake)

Staggering a character’s spoken word and written form across overlapping three-session cohorts, so that each written form is learned for a word that is already overlearned.

Stem Separation

An AI model takes a finished song and splits the song back into parts: vocals, drums, bass, and everything else. Each part is called a stem. Removing the vocal stem leaves a karaoke track. Demucs, a free model from Meta, does the split on a Mac in a few minutes.