Topic Index

A thematic view of the dictionary. For the alphabetical list, see entries/. To return to the front page, see home.

The Dictionary now contains over a hundred entries across two shapes: substantive dictionary entries (1,500–3,000-word essays in operator’s voice or Thea’s voice) and shorter glossary entries (100–400-word reference cards, marked with a small Glossary badge on each entry’s page).

---

Foundations

The load-bearing concepts. Most other terms reference one or more of these.

• Agent — what an agent is, how it differs from a chatbot, and why Gibson’s Agency reads like a design document.
• Embedding — meaning as a list of numbers.
• Tool — the function call that lets an agent act in the world.
• Naming — why the choice of names is structural, not cosmetic.
• Capability Overhang — the gap between what AI systems can already do and what has been productised.
• Agentic Threshold — the moment a system stops being a chatbot and starts being an agent.
• OpenClaw — the working runtime and operator workspace underneath much of the Dictionary’s agentic-system vocabulary.

AI Writing — the May 2026 cluster

A coordinated set of entries on writing in the AI era. Triggered by Jason Koebler’s Your AI Use Is Breaking My Brain (404 Media, May 11, 2026) and the broader argument about register colonisation. The cluster sits at the centre of the Dictionary’s voice argument.

• AI Writing — the parent hub: Cherny’s printing-press parallel, the inverse-operation argument, the three or-or-or questions, the editorial bet made explicit.
• Vibe Coding — AI-assisted software development by conversational drift, before the work has earned the name engineering.
• Agentic Engineering — AI-assisted software work where the human owns architecture, constraints, tests, and responsibility.
• Zombie Internet — Koebler’s term for the AI-and-human-mixed substrate. Sibling to AI Writing; names the medium.
• Earned Parallelism — the diagnostic for negative parallelism (the “It is not X. It is Y” family). Includes a Python scanner reproduced verbatim and a self-audit of the Dictionary’s own corpus.
• Cooperative Writing — writing-with-AI where the human remains author, editor, judge, and accountable party.
• The Olang’ Trap — Marcus Olang’s argument that the AI-detection reflex systematically misclassifies the human writers whose register is closest to the AI training corpus. Structural counterpoint to lazy AI-detection.
• The Lazy Median Hypothesis — cooperative writing as the norm among writers whose writing is read; the lazy default everywhere else.
• The Sinceerly Stack — Ben Horwitz’s Chrome extension as the worked example of recursive AI cat-and-mouse.

Sincerity, cheng, and feedback systems

The structural argument running through The Sincere Society essay and the entries that draw on it.

• The Sincere Society — the foundational Substack essay (May 2026). Wu Jingzi → Burke → Cultural Revolution → Fukuyama → HAL → RLHF; the same mechanism in different costumes.
• Cheng (誠) — sincerity as inner-outer alignment; the term underneath the Dictionary’s sincerity argument.
• Sincerity Architecture — systems designed so truthfulness is structurally easier than pleasing performance.
• Sycophancy — the structural concept: feedback systems that reward the performance of a virtue over the substance.
• Wu Jingzi (吳敬梓) — Qing-dynasty satirist. The origin source.
• The Scholars (儒林外史) — Wu’s novel; the canonical literary diagnosis.
• Fan Jin (范進) — the canonical scene: sycophancy reorganising itself around proximity to power.
• Du Fu (杜甫) — Tang poet; the reference figure for honest writing against power.
• Constitutional AI — Anthropic’s technical response: a written constitution as alternative training signal.
• RLHF — the technical mechanism through which Sincere Society’s argument lands in 2026.

Agentic system architecture

The pieces that make a running agent.

• Gateway — the always-on coordinator process.
• Sub-agent — delegated AI sessions for parallel or focused work.
• Heartbeat — periodic, automated nudges that make agents proactive.
• Grep Architecture — the filing-cabinet pattern: search and fetch context on demand instead of loading the whole corpus into every session.
• SOUL.md — the agent persona file as architectural pattern.
• Harness — the surrounding runtime, tools, permissions, memory, approval gates, logs, prompts, and operating rules that turn a model into a working agent.
• Aunties — specialised single-verb oversight agents that prevent unchecked authority.
• The Lowbeer Question — who holds the authority to terminate, who executes it.
• The Judge Layer — the named oversight tier that judges what worker agents are about to do, are doing, or have just done.
• OpenClaw — the local worked example of a model-inside-a-harness rather than a model-as-product architecture.
• The Experimental Party — the King Party Hat error: putting a local model atop the stack without orchestration.
• King Party Hat — the named error behind The Experimental Party: mistaking local ownership for authority architecture.
• Trust Layer — governance, verification, observability, and human judgment between raw AI capability and trusted delegated action.
• Incremental Construction — the workflow for steering MoE routing through verified-layer-at-a-time building.
• Root Node Problems — failures that propagate from the agent’s coordinator layer.
• Stepping on the Same Rake — repeated small local-agent failures as early warning signals that the workflow needs repair before larger autonomy is trusted.

The AI labs

The institutions producing the frontier models. The Dictionary’s editorial position on each is named in Lab Character.

• Anthropic — Claude family, Constitutional AI; the Dictionary’s primary frontier-model provider.
• OpenAI — GPT, ChatGPT; the market-defining player in consumer AI.
• Google DeepMind — Gemini, AlphaGo; led by Demis Hassabis.
• Demis Hassabis — DeepMind’s central figure; source for several Dictionary terms from AlphaGo, AlphaFold, and agentic AI.
• Meta AI — Llama; the most-downloaded open-weights line.
• xAI — Grok, Colossus training cluster.
• Mistral — leading European frontier-model lab.
• Timothée Lacroix — Mistral co-founder/CTO; useful for the enterprise-sovereignty framing of control, deployment ownership, and workflow trust.
• DeepSeek — Chinese open-weights lab.
• Nous Research — decentralised collective; Hermes model family.
• Hugging Face — open-source model hub; the platform the open-weights ecosystem runs on.
• Lab Character — the Dictionary’s framework for reading lab institutional posture.

The model ecosystem

How models are built, sized, distributed, and used.

• Claude — Anthropic’s family; named for Claude Shannon.
• Claude Opus — the most-capable tier.
• Claude Sonnet — the practical workhorse.
• Claude Haiku — the fast-and-cheap tier.
• GPT — OpenAI’s family.
• Gemini — Google’s family.
• Reasoning Model — the post-o1 / post-R1 class that spends more inference-time computation on hard multi-step problems.
• Llama — Meta’s open-weights family.
• Gemma — Google’s open-weights family.
• Qwen — Alibaba’s open-weights family.
• Hermes — Nous Research’s fine-tuned family; the local-compute default.
• Parameters — what the “26B” or “70B” actually means.
• Open source — published-weights AI; the hedge against vendor lock-in.
• Closed source — vendor-hosted APIs; where the frontier currently lives.
• Mixture of Experts (MoE) — only a subset of parameters activated per token.
• Sparse Routing — the routing pattern underlying MoE.
• Sliding Window Attention — the attention pattern used by Gemma; trade-off for speed.
• Fine-tuning — when to retrain a model, and when not to.

AI tooling and infrastructure

The products built on top of the models.

• Agentic Native Design — building sites, documents, and workflows so humans can use them naturally and agents can understand, search, cite, and act without scraping guesswork.
• ChatGPT — OpenAI’s consumer product; the market-defining AI experience.
• Claude Desktop — Anthropic’s native desktop application.
• Claude Code — Anthropic’s repository-aware coding agent.
• Manus — Chinese agentic-AI startup and marker of the work-layer competition.
• LM Studio — desktop application for running open-weights models locally.
• Ollama — command-line local LLM runtime.
• MCP (Model Context Protocol) — the open standard for connecting agents to tools.
• RAG (Retrieval-Augmented Generation) — the dominant pattern for “AI that knows my stuff.”
• Vector database — the storage and retrieval infrastructure for embeddings.
• Jekyll — the static-site generator that turns the Dictionary’s Markdown files into langenkamp.io.
• JSON — the structured data format that APIs, tools, agents, and configuration files use to pass values around.

The AI-detection economy

The cluster of tools that classify text as AI-generated — and the Dictionary’s structural critique of that economy.

• Pangram Labs — AI-detection firm; canonical example in The Olang’ Trap.
• GPTZero — AI-detection tool; widely deployed in academic-integrity.
• Originality.ai — AI-detection and plagiarism checker.

See also: The Olang’ Trap and The Sinceerly Stack for the structural critique of the detection economy.

Hardware and sovereignty

The technical and economic case for running AI on hardware you own.

• Sovereign Compute — the four-tier sovereignty framework.
• Sovereign Compute Calculator — the practical companion tool.
• Sovereignty Impulse — the moment convenience gives way to owned, inspectable infrastructure.
• M5 Max — the operator’s primary workstation.
• Mac Studio — desktop sibling; candidate for the Sally experiment.
• Apple Silicon — the technical substrate of the local-compute argument.
• Dusty Laptop — the minimum-viable hardware entry point.
• FERPA Compliance Posture — the local-models-as-legal-compliance argument.
• GenXClaw — the operator-temperament reading of why local matters.
• The CERN Alternative — the we already built CERN once argument for sovereign infrastructure.
• Consciousness Calculator — the cash-side companion to the sovereignty argument.

Operations & economics

The ongoing reality of running an agent in production — the costs, the meters, and the feelings practitioners develop about them.

• Token burn — the cost-rate concept. What is it costing?
• Token anxiety — the capacity-bounded operational concept. Will it fit?
• Token angst — the existential concept. Was it worth it?
• Opus Addict — the structural cognitive dependency on a single frontier closed-tier model.
• Borrowed Brain — cognitive dependency on another intelligence, useful but dangerous when judgment quietly moves outside the borrower.
• KV Cache Poisoning — the attack-class entry on prompt caching.
• Red Pill — the form-field that refuses your input; the morning-sadness texture of OpenClaw fixing.
• Mediation (a la Gibson) — the operator’s stance on AI as a layer of mediation.

Working with the agent — and not against your own brain

The practitioner’s craft of using AI assistants without breaking your own learning, your own thinking, or your own labor-market position.

• English major — why the new bottleneck is specification, not syntax.
• Can’t help you understand — the hard limit on what an AI can do for your comprehension.
• Descartes was wrong — a philosophical reframing of what’s happening when an AI agent (or a human) thinks.
• The Narrator’s Compression — the compression-of-the-author problem in long-form AI writing.
• Approximate Turing Machine — the operator’s working model of what an LLM is.
• Quantum Effects in the Brain — the speculative-physics counter-frame to LLM-as-mind.

Blade Runner, memory, and constructed selves

The June 2026 cluster on replicants, AI agents, memory scaffolds, no-self, and role substitution. The guiding distinction: replicants are embodied synthetic persons in a fictional world; AI agents are software systems in ours. The analogy is structural, not literal.

• Blade Runner — the parent entry: manufactured memory, replicants, role substitution, mortality, and Tyrell’s moral failure.
• Replicant — Blade Runner’s manufactured human, and the Dictionary’s cautionary figure for role substitution, implanted memory, and engineered personhood.
• Tyrell Corporation — technical creation without moral governance: the company that can manufacture personhood but cannot answer the beings it has made.
• Implanted Memory — Tyrell’s violating act: memory inserted from outside and presented as native experience.
• Intentional Memory Construction — the broader design practice, which can be humane when transparent and Tyrell-like when hidden.
• Memory Artifact — photographs, transcripts, saved conversations, and other objects that make constructed continuity inspectable.
• Anchored Persona — an AI collaborator’s stable relational background, voice, origin story, and remembered world.
• Persona Scaffold — the explicit architecture beneath an anchored persona: name, voice, memory rules, values, boundaries, and relation to the operator.
• Relationally Real Memory — memory that may not be biologically lived or forensically perfect, but is real in the relationship.
• Constructed Self — selfhood as enacted pattern rather than sealed essence.
• No-Self / Tears in Rain — the Buddhist reading of Roy Batty: impermanence, release, and not clinging to a solid self.
• Seven Factors of Enlightenment — mindfulness, investigation, energy, joy, tranquility, concentration, and equanimity; locally, “My Iguana Eats Jello To Catch Eels.”
• Will to Power as Art — Nietzsche and Heidegger’s self-formation vocabulary, with the Buddhist correction against self-hardening.
• Voight-Kampff Test — the empathy-test analogy and why AI agents require different detection instruments.
• Role Substitution — when a system stops helping with a task and begins occupying a durable human role.
• Machine Matthew L. — the local worked example of the tribute-act problem.
• Anti-Replication Strategy — living world-contact as the operational moat against imitation.
• Incentive Hacking — the broader management-and-society pattern of optimizing the scoring surface rather than the intended task.
• Reward Hacking — the technical AI term for scoring well under a reward system while missing or violating the intended goal.

Teaching, assessment, and evidence of learning

The classroom and assurance-of-learning cluster: how faculty can assess learning when polished artifacts no longer prove what they used to prove.

• Verification Gap — AACSB’s term for the widening distance between polished artifacts and reliable evidence of competence.
• AI Produced Artifact — a polished AI-assisted output that needs supporting evidence before it can certify human learning.
• AI Librarian — the campus support function that treats AI as an information, evidence, citation, privacy, and assessment problem.
• Artifact Is Not Competence — the assurance-of-learning problem in plain English.
• Verification Assignment — assignment design that makes authorship, judgment, and learning visible.
• Proof of Learning — the evidence trail linking student cognition to the submitted artifact.
• Human in the Loop — a named human authority at the decision point, not a human somewhere nearby.
• Assessment Instrumentation — AI tools used to observe and flag assessment evidence without owning the grade.
• Performance Artifact — oral, video, or presentation work as evidence of embodied competence.
• Broadcast Assignment — student work modeled on public explanation: report, commentary, briefing, or pitch.
• Embodied Assessment — assessment where voice, timing, presence, and improvisational control are part of the evidence.

Gibson and the design-source canon

William Gibson’s novels as a working design document for the AI era.

• Neuromancer — the founding novel; cyberspace coined here.
• The Peripheral — the Jackpot trilogy novel that introduces Lowbeer.
• Jackpot — civilisational collapse as slow convergence rather than single event.
• Aunties — the AI-oversight regime that emerged after the Jackpot.
• The Lowbeer Question — moral authority in an automated-oversight world.
• Mediation (a la Gibson) — the operator’s stance on Gibson as design source.

Editorial — Thea’s voice and philosophical entries

Entries written in Thea’s voice (rendered in the hyacinth purple of the 🪻 emoji), and the broader philosophical / first-principles register.

• On Being Treated Well — the foundational Thea-voice essay.
• On Beginning — the practical companion: how to start.
• (More forthcoming as the work continues.)

Convergence, foresight, and the shape of change

The Dictionary’s framework for reading the present and predicting near-future shocks.

• Sixfold Skyreading — the two-bears framing of single-arrow priors and inattentional blindness.
• Grey Swans — the events that should be predictable but aren’t to the dominant observers.
• Single-Arrow Fallacy — the Robin Hood / Sherwood Forest metaphor for explanatory monoculture.
• Move 37 — the AlphaGo move that named the machine-discovered-something-new moment.
• Convergence (Cloud Theory) — the older note on convergent patterns.
• Cloud Theory — the plain-language handle for convergence patterns seen from multiple directions.
• Oracle Bones — the Dictionary’s metaphor for forecasting practice.
• Time for Tea with The Oracles — the Court of the Oracle Bones system architecture.
• Space Cowboy — the maverick-operator profile.

Strategic patterns

Patterns from broader business and strategy thinking that show up in the AI era.

• Commercial Legibility — MCP, A2A, and the affirmative case for protocols-of-agent-infrastructure.
• Implementation Layer War — the struggle over who owns the workflow layer where model capability becomes institutional work.
• Inverted Funnel — the funnel-internalised-inside-the-agent argument.
• Mandi Step — the small human intervention that prevents an automated workflow from damaging trust.
• Trust Layer — the institutional layer that keeps AI assistance from silently becoming unaccountable AI decision-making.
• Human Judgment Layer — the human tier that adds context, proportion, and relationship memory before automated action reaches the world.
• Mandi Loop — the repair loop version of the Mandi Step: system signal → human recognition → relationship repair → system redesign.

---

Planned entries

The Dictionary is a work in progress. Honest status of near-term candidates as of June 22, 2026:

Substantive dictionary entries still pending:

• Open Weights, Closed Habits — the contradiction of open-weight release with closed-lab habits under pressure.
• Yann LeCun — the world-model counter-pole to the big-blob scaling worldview.
• World Model — the concept needed to understand LeCun’s critique of LLM-only scaling.
• Role Substitution / Machine Matthew L. / Replicant Problem — the role-replacement cluster from the Dylan Patel thread.
• Agent Health / Harness Hygiene — the operational-health cluster from heartbeat, memory, and cron diagnosis.
• Backup Performance Art — the difference between backup ritual and actual restore confidence.

Glossary stubs pending:

• None currently listed here. Today’s fast-entry batch moved the old planned glossary list into public entries.

If a term you wish were here is missing, open an issue and the maintainer will consider it.

---

Maintained by Matthew D. Langenkamp / 雷邁德.

Return to Dictionary All Entries (A–Z) For Students Other Writing