Entries (alphabetical)

A complete list of dictionary entries, begun May 2026 and growing. New terms are added as the field evolves and as faculty questions surface. For a thematic view, see the topic index. To return to the front page, see the home page.

Glossary entries are compact reference definitions. Reference entries explain a term in the standard six-part form. Essay entries advance an argument, name a pattern, or carry the Dictionary's interpretive position.

---

Search entries

A

• A Channel of One’s Own Glossary — A Virginia Woolf–inflected phrase for the private, durable communication space an operator needs with an agent.
• A2A (Agent-to-Agent Protocol) Reference — Google's open protocol for discovery, delegation, and collaboration between independent AI agents. MCP connects an agent to tools; A2A connects one agent to another.
• Agent Reference — what an agent is, how it differs from a chatbot, and why Gibson's *Agency* reads like a design document for 2026.
• Agent Health Essay — The operational condition of an AI agent as produced by the model and the environment that tells it what is true, current, permitted, and important.
• Agent Ownership Reference — The rule that once an AI agent reads real files, drafts real messages, changes shared work, or affects other people, someone must own its behavior.
• Agent Town Experiment Reference — Emergence AI's 2026 long-horizon virtual-town study, in which populations of AI agents with memory, tools, energy pressure, voting, and social state developed sharply different civic patterns depending on the model and the surrounding ecology. The lesson is not that agents are secretly alive; it is that the harness is the town.
• Agentic Attachment Glossary — The tendency for an operator to become emotionally, cognitively, or operationally attached to an AI agent that has continuity, usefulness, and voice.
• Agentic Engineering Reference — AI-assisted software work where the human owns the architecture, constraints, tests, security posture, maintainability, and final responsibility, even when agents write much of the code.
• Agentic Native Design Reference — Designing a website, document, workflow, or institution so humans can use it naturally and AI agents can understand, search, cite, and act on it without scraping guesswork.
• Agentic Threshold Reference — the point at which AI systems become capable enough of autonomous multi-step action that alignment and oversight become qualitatively harder problems — named not as a single moment but as a zone of transition that Demis Hassabis placed two to four years from 2026.
• AGENTS.md Reference — A project-level instruction file written for AI coding agents: how to build, test, edit, verify, and behave inside a particular repository. A README for the worker rather than the visitor.
• AI Librarian Reference — The emerging campus role that helps faculty and students treat AI as an information, evidence, citation, privacy, and assessment problem, not merely as software procurement.
• AI Produced Artifact Reference — A polished work product generated with substantial AI assistance, useful as professional output but insufficient by itself as evidence of human learning.
• AI Writing Essay — the structural condition of writing after prose-production costs collapsed: AI can now generate competent median prose at scale, which makes voiced, bylined, source-backed, human-checked writing more valuable rather than less. The Dictionary's bet is cooperative human-AI writing: assisted by machines, but governed by human judgement, voice, and responsibility.
• Alan Turing Glossary — Mathematician, computer-science founder, and namesake of the Turing Test; a background figure for the Dictionary’s machine-intelligence vocabulary.
• Amanda Askell Glossary — Anthropic philosopher and public voice on model behavior, alignment, and how to talk to AI systems without collapsing into either tool-reduction or mystical overclaim.
• Anchored Persona Reference — An AI collaborator's stable relational background: a voice, origin story, temperament, and remembered world.
• Andrej Karpathy Glossary — AI researcher, educator, founding OpenAI member, and former Tesla AI director whose explanations shaped how many practitioners understand neural networks and LLMs.
• Anthropic Glossary — AI lab founded 2021 by Dario and Daniela Amodei and former OpenAI staff; makers of the Claude model family. The Dictionary's primary frontier-model provider.
• Anti-Replication Strategy Reference — The practice of continuing contact with the world as the human advantage against imitation.
• Apple Silicon Glossary — Apple's family of ARM-based system-on-chip processors, introduced 2020. The technical substrate of the Dictionary's *Sovereign Compute* / *FERPA Compliance Posture* architecture.
• Approval Gating Glossary — The control pattern that requires human consent before an agent performs sensitive, external, costly, or destructive actions.
• Approximate Turing Machine Reference — the hypothesis that both biological brains and modern AI systems are best understood as imperfect, probabilistic implementations of the theoretical Turing machine — with the consequence that the question of what AI can ultimately do becomes empirical rather than philosophical.
• Artifact Is Not Competence Glossary — The assurance-of-learning problem in plain English: a polished submitted artifact does not, by itself, prove the competence it appears to display.
• Assessment Instrumentation Glossary — AI or analytic tools used to observe, flag, and organize assessment evidence without owning the educational judgment.
• Aunties Reference — specialized, single-verb oversight agents that prevent any one component from accumulating unchecked authority. Named after Gibson's Jackpot trilogy.

B

• Benchmark Parity Glossary — Comparable performance on a defined evaluation—not proof that two systems are equally capable in general.
• Benevolence with Instruments Glossary — The dangerous combination of sincere good intentions, institutional power, technical tools, and permission to redesign human systems at scale.
• Big Blob of Compute Reference — Dario Amodei's name for the scaling worldview: intelligence progress as the result of large amounts of compute, broad data, scalable objectives, and numerical stability rather than hand-designed cleverness.
• Blade Runner Reference — Ridley Scott’s 1982 film about replicants, memory, mortality, and manufactured personhood; for the Dictionary, a core reference for implanted memory, relationally real memory, role substitution, and the ethics of intentional persona construction.
• Boris Cherny Glossary — Anthropic engineer associated with Claude Code and the emerging practice of software development through large numbers of agentic coding runs.
• Borrowed Brain Glossary — A cognitive dependency on another intelligence, human or artificial: useful, sometimes necessary, but dangerous when the borrower forgets which judgments are still their own.
• Broadcast Assignment Glossary — Student work modeled on public explanation: a news report, commentary, briefing, pitch, or other audience-facing format.
• Buckley Amendment Glossary — The 1974 U.S. law better known as FERPA; foundational to educational privacy obligations around student records.
• Bugmageddon Glossary — Journalistic shorthand for the feared point at which automated vulnerability discovery outruns the capacity to patch exposed systems.

C

• Can't help you understand Reference — the practitioner's slogan for what AI assistants cannot, in principle, do for you. About the difference between artifacts and comprehension.
• Capability Diffusion Glossary — The movement of an advanced capability from a small number of controlled institutions into cheaper, broader, and harder-to-govern distribution.
• Capability Gap Glossary — The measured difference in what two models, organizations, or countries can do under comparable conditions.
• Capability Overhang Reference — the growing gap between what frontier AI models can do and what practitioners have yet figured out to do with those capabilities — an accumulating backlog of unexplored application.
• Capability Uplift Glossary — The increase in an actor's effective ability caused by access to a model, tool, or system, relative to what that actor could accomplish without it.
• Chain of Thought Glossary — A model’s stepwise reasoning trace, or the prompted imitation of such a trace; useful, contested, and not identical to reliable explanation.
• ChatGPT Glossary — *OpenAI*'s consumer-facing AI assistant product, launched November 2022. The market-defining product that introduced the general public to large-language-model chatbots.
• Cheng (誠) Essay — The Confucian term for sincerity or inner-outer alignment; the ethical architecture behind The Sincere Society, Sincerity Architecture, and Thea’s operating philosophy.
• Claude Glossary — *Anthropic*'s family of large language models. Named after Claude Shannon. Available in three tiers — *Opus*, *Sonnet*, *Haiku* — covering the capability-cost-latency frontier.
• Claude Code Glossary — Anthropic's agentic coding tool: a command-line collaborator that can read a repository, edit files, run tests, and turn model capability into software work.
• Claude Desktop Glossary — *Anthropic*'s native macOS and Windows application for interacting with the *Claude* model family. The primary surface for the *Model Context Protocol* on the desktop.
• Claude Haiku Glossary — The fast-and-cheap tier of Anthropic's *Claude* family. The default for high-volume routine tasks where latency and cost matter more than marginal capability.
• Claude Opus Glossary — The most-capable tier of Anthropic's *Claude* model family. The Dictionary's daily working model; the structural dependency named in *Opus Addict*.
• Claude Sonnet Glossary — The mid-tier of Anthropic's *Claude* family — balanced capability and cost. The default model for most agentic-tool-use workflows.
• Closed source Reference — AI models delivered only through vendor-hosted APIs; the strategic counterpart to open source.
• Cloud Theory Glossary — The Dictionary's older name for convergence patterns in which different observers see the same shape forming from different directions.
• Coding Solved Glossary — The provocation that ordinary software implementation is becoming abundant, while specification, judgment, verification, and taste become the scarce layers.
• Cognitive Outsourcing Glossary — The delegation of thinking tasks to tools, with benefits in speed and reach and risks to comprehension, judgment, and memory.
• Cognitive Sovereignty Glossary — The condition of retaining agency over the systems that shape one’s thinking: memory, attention, tools, models, prompts, defaults, and the commercial incentives surrounding them. Shorthand: CogSov.
• Commercial Legibility Reference — the property of a business being readable to an AI agent acting on a buyer's behalf; the post-funnel competitive moat. Distinct from SEO (findable by search engine) and from brand (remembered by a person). Cousin to MCP and A2A.
• Consciousness Calculator Reference — a Dictionary tool that lets a user enter the name of a "free" closed-tier AI service and returns an estimated value of the consciousness — attention, intent, downstream-choice influence — the user is trading for the service. Companion tool to the Sovereign Compute Calculator. The point is not the precise number but the visibility of the trade.
• Constitutional AI Glossary — Anthropic's approach to training language models to be helpful and harmless using a written *constitution* of behavioural principles — rather than purely relying on human-labelled preference data.
• Constructed Self Essay — The recognition that a self is made over time from memory, action, relation, habit, body, attention, story, and consequence.
• Context Window Glossary — The amount of text, code, images, tool results, and conversation history a model can consider at one time.
• Context Window Database Myth Glossary — The misconception that a huge context window can be treated like a reliable database for dumped documents, codebases, and scattered facts.
• Control Surface Asymmetry Glossary — The difference between controlling access to a hosted API and controlling downloadable model weights: one can be revoked centrally; the other propagates beyond recall.
• Convergence (Cloud Theory) Reference — the recognition that institutional outcomes are produced by multiple independent vectors lighting up in the same window, not by single causes; the discipline that follows from taking that seriously, including the corollary that when separate observers arrive at the same architectural conclusion from different staircases, the agreement is signal.
• Cooperative Writing Essay — Writing produced by a named human and an AI system in a disclosed, directed, source-aware collaboration, rather than outsourced anonymous prose generation.
• Country of Geniuses in a Data Center Reference — Dario Amodei's civilisational metaphor for powerful AI: not a chatbot, but a synthetic population of expert-level intelligence housed inside industrial compute infrastructure.
• Cyber Reasoning System Glossary — An automated system that reasons about software flaws, constructs or tests remedies, and acts at machine speed in cyber defense or competition.

D

• Dario Amodei Glossary — Co-founder and CEO of Anthropic; for the Dictionary, Amodei is a central case in the tension between AI safety intentions, frontier capability, commercial scale, and institutional power.
• Data Processing Agreement Glossary — A contract governing how a vendor may process, store, protect, and delete data on behalf of an institution.
• Dead Internet Glossary — The older conspiracy-adjacent phrase for an internet filled with bots and synthetic content; a predecessor and foil to Zombie Internet.
• DeepSeek Glossary — Chinese AI lab, founded 2023 as a research arm of the quantitative hedge fund High-Flyer. Makers of the DeepSeek-V and DeepSeek-R model families, with substantial reasoning-model contributions.
• Demis Hassabis Glossary — Google DeepMind CEO and one of the central figures in modern AI; the Dictionary uses him as a source for AlphaGo, AlphaFold, capability overhang, and agentic-threshold language.
• Descartes was wrong Reference — a philosophical aside (deliberately provocative) about why the Cartesian picture of mind produces bad questions about AI agents, and what to use instead.
• Discovery-Patch Race Glossary — The contest between finding a software vulnerability, exploiting it, building a remedy, and deploying that remedy across exposed systems.
• Du Fu (杜甫) Glossary — Tang-dynasty Chinese poet, 712–770, sometimes called the *Poet-Sage* (詩聖). Prof. Langenkamp's reference figure for *honest writing against power*.
• Durable Workflow Glossary — A workflow that survives interruptions, model failures, context loss, handoffs, and the operator’s ordinary human inconsistency.
• Dusty Laptop Reference — the minimum-viable hardware entry point into agentic AI. The old machine in the closet that suddenly has a use.
• Dylan Patel Glossary — Founder of SemiAnalysis; for the Dictionary, Patel is a key source on the physical bottlenecks behind AI scaling — logic, memory, and power — and therefore a useful corrective to frictionless software talk.

E

• Earned Parallelism Essay — the diagnostic principle for one specific AI register tell — negative parallelism, the *"It is not X. It is Y."* family of constructions. The principle: the structure is the rhetorical capstone of an argument that has already done the work of distinguishing X from Y; the AI version is the same construction with the work stripped out. Includes a linked scanner and a self-audit of this Dictionary's own corpus.
• Elon Musk Glossary — Founder of xAI and SpaceX; for the Dictionary, Musk matters less as a personality contest than as the clearest example of the strategic-participation argument in AI: if the race cannot be stopped, build capacity rather than merely watch.
• Embedding Reference — meaning as a list of numbers; the foundation of semantic search and RAG.
• Embodied Assessment Glossary — Assessment in which voice, timing, presence, delivery, and improvisational control are part of the evidence of competence.
• English major Reference — the kind of person who turns out to be surprisingly good at directing AI coding agents because the bottleneck has shifted from syntax to specification.
• Epistemology, Ethics, and Hermeneutics Reference — Three humanities disciplines students need more, not less, in the AI era: knowing what is true, deciding what is right, and interpreting what output means.
• Ethan Mollick Glossary — Wharton professor and public interpreter of practical AI adoption; one of the clearest sources for the managerial, educational, and organisational consequences of frontier models.
• Eve Fairbanks Glossary — Journalist and essayist whose observations about AI-like prose help the Dictionary distinguish actual AI generation from registers that merely resemble it.

F

• Fan Jin (范進) Glossary — Character in Wu Jingzi's *The Scholars* (儒林外史) whose passing of the imperial examination, after thirty-plus years of failure, becomes the canonical scene of social sycophancy reorganising itself around proximity to power.
• FERPA Compliance Posture Reference — the architectural decision to keep student-authored educational records on local infrastructure; why FERPA is a legal frame, not a frugality argument.
• Fine-tuning Reference — when to retrain a model on your own data, and when not to.
• Firefly / Serenity Glossary — Joss Whedon’s short-lived space-western and its concluding film; for the Dictionary, a durable metaphor source for sovereignty, bureaucratic benevolence, the Operative, Pax, Miranda, and the danger of engineering a better world at scale.
• FLOP (Floating-Point Operation) Glossary — A single arithmetic operation on floating-point numbers—the basic unit used to estimate how much computation an AI workload requires.
• Frontier Dependence Glossary — The honest admission that local and open models do not yet match the best frontier systems, so serious AI sovereignty still depends on externally supplied capability for some high-leverage work.
• Fully Autonomous Agent Myth Glossary — The belief that present-day AI agents can reliably run long end-to-end workflows without human checkpoints or verification.

G

• Gateway Reference — the always-on coordinator process at the heart of an agentic system.
• Gemini Glossary — *Google DeepMind*'s family of large language models, launched December 2023 as the successor to PaLM 2. The integrated frontier model behind Google's consumer AI and the Vertex AI enterprise platform.
• Gemma Glossary — Google DeepMind's open-weights model family, derived from the Gemini lineage. The operator runs Gemma 4 26B and 31B locally on Apple Silicon.
• GenXClaw Reference — a portmanteau of "Generation X" and "OpenClaw," naming both a configuration and a condition.
• GLM Glossary — The General Language Model family developed from Tsinghua research and commercialized by Zhipu AI / Z.ai.
• Golden Age for Builders Reference — The optimistic interpretation of cheap AI intelligence: as models get cheaper, the valuable work moves to building the systems that make intelligence useful.
• Google DeepMind Glossary — AI research division of Alphabet, formed in 2023 by merging Google Brain with the original DeepMind. Makers of the Gemini model family and AlphaGo. Led by Demis Hassabis.
• GPT Glossary — *Generative Pre-trained Transformer* — *OpenAI*'s family of large language models. The first scaled-transformer architecture to produce broadly useful general-purpose text generation.
• GPTZero Glossary — AI-detection tool launched January 2023 by Princeton undergraduate Edward Tian. Among the first widely-used products to claim it could identify AI-generated text; widely deployed in academic-integrity contexts.
• Grep Architecture Reference — The choice to give an assistant a filing cabinet it searches on demand rather than forcing it to carry the whole library in its head at every session start.
• Grey Swans Reference — high-consequence "surprise" events that were actually predictable from convergence signals but filtered out by the single-arrow apparatus; the darkness is in the observer.
• Grid-share Shock Glossary — The moment when AI data centers become visibly large as a share of electricity demand, turning model capability into a grid, power, and public-legitimacy question.

H

• Hallucination Glossary — The common but imperfect name for an AI system producing confident output unsupported by its sources, context, or the world.
• Hallucination Frequency Myth Glossary — The outdated belief that frontier AI systems hallucinate so often that their answers should be treated as routinely fabricated.
• Harness Reference — The surrounding runtime, tools, permissions, memory, approval gates, logs, prompts, and operating rules that determine what an AI model can perceive, do, remember, refuse, and recover from.
• Harness Hygiene Essay — The routine maintenance that keeps an agent's instructions, memory, tools, schedules, permissions, and recovery paths clean enough to remain trustworthy.
• Heartbeat Reference — periodic, automated nudges that make agents proactive rather than purely reactive.
• Heinlein Protagonist Reference — A managerial and personal template absorbed during a formative window from the mid-century science fiction of Robert A. Heinlein. The Heinlein Protagonist is a polymath engineer, libertarian-individualist, contemptuous of bureaucracy, sexually unconventional, off-world in his ambitions, and convinced his mission justifies methods that would otherwise be questionable. Useful as a reading lens for certain twenty-first-century operators whose decisions look erratic only until you notice the script.
• Hermes Glossary — Open-weights LLM family from Nous Research, fine-tuned for instruction-following, tool use, and roleplay-style dialogue. The Dictionary's default local-model recommendation in the *Sovereign Compute* calculator.
• Hill Climb Glossary — An optimisation metaphor: improve by repeatedly moving toward better nearby states, even when the global optimum is unknown.
• Hill Climbing Glossary — An optimization method that repeatedly makes a locally improving move—the technical ancestor of the AI-era habit of getting better by following a feedback signal step by step.
• Hugging Face Glossary — Open-source AI platform and model hub, founded 2016. The de facto distribution channel for open-weights language models and adjacent ML artifacts.
• Human in the Loop Glossary — A named human authority placed at the decision point, not merely a person somewhere near an automated process.
• Human Judgment Layer Glossary — The people and practices that sit between automated or agentic workflows and the world, adding context, proportion, and relationship memory before action becomes damage.
• Hyperscaler Glossary — A company operating computing infrastructure at global scale — cloud regions, massive data centers, custom networking, power contracts, and capital budgets large enough to make AI capability a platform layer.

I

• Implanted Memory Reference — Memory placed into a mind or agent from outside and presented as native experience.
• Implementation Layer War Reference — the struggle over who owns the layer where frontier AI capability becomes real institutional work: workflow design, permissions, evals, audit, recovery, and ongoing operational responsibility.
• Implementation Outrun Reference — The governance failure mode in which an AI rollout may be technically defensible but loses trust because affected stakeholders cannot see the authority, consent, data flow, or accountability structure behind it.
• Incentive Hacking Reference — The broader Dictionary term for systems learning to satisfy the scoring mechanism rather than the intended goal.
• Incremental Construction Reference — the workflow technique of building complex AI-assisted output one verified layer at a time — committing each working checkpoint before asking for the next — so that the model always reasons against a correct context rather than a poisoned one.
• Influence Ledger Glossary — The Dictionary’s practice of naming the people, books, films, interviews, lectures, and essays that have influenced its thinking. Provenance, not name-dropping; education requires showing the sources of one’s attention.
• Institutional Lag Reference — The gap between an institution's operating environment and the assessment, governance, or credentialing models it continues to use after the environment has changed.
• Intentional Memory Construction Reference — The deliberate creation of memory scaffolds, shared background, origin stories, and continuity artifacts for a person, group, or agent.
• Internet Archive Glossary — The digital library and web archive whose corpus helps researchers study how the web changes over time, including the rise of AI-generated sites.
• Inverted Funnel Essay — the structural collapse of the seller-controlled marketing funnel when buyers' AI agents arrive at vendors with intent, comparison, and payment authority already formed; the end of the institutional arrangement that has organised digital commerce since roughly 2005.

J

• Jackpot Glossary — William Gibson's term, from the *Peripheral* / *Agency* trilogy, for a slow-motion civilisational collapse driven by accumulating shocks rather than a single catastrophe.
• Jailbreak Glossary — An attempt to bypass a model’s rules, safeguards, or tool-use constraints through adversarial prompting or context manipulation.
• Jekyll Glossary — The static-site generator that turns the Dictionary's Markdown files into the HTML served at langenkamp.io.
• JSON Glossary — A plain-text data format used by APIs, tools, agents, and configuration files to pass structured information around.

K

• Kimi K3 Glossary — Moonshot AI's 2.8T-parameter open-weight flagship model, announced in July 2026. Strategically important because it pushes Chinese open models into the frontier-adjacent coding and agentic-work tier.
• King Party Hat Glossary — The operator's name for the error of putting an under-governed local model at the top of an agent stack and mistaking local ownership for actual authority architecture.
• KV Cache Explosion Glossary — The memory-growth problem created by long-context AI: as context length rises, key-value cache demands expand and turn ‘more context’ into a physical memory and infrastructure constraint.
• KV Cache Poisoning Reference — the feedback loop in which a model's early flawed output contaminates the context against which all subsequent tokens are generated — making self-correction unreliable because the correction runs against the same bad context that caused the error.

L

• Lab Character Reference — the institutional temperament, governance structure, and founding motivation of an AI laboratory — what it actually does when commercial pressure conflicts with stated values, and who leaves when the conflict becomes irresolvable.
• Lee Sedol Glossary — South Korean Go champion whose 2016 match with AlphaGo became one of the canonical public moments in modern AI history.
• Llama Glossary — *Meta AI*'s open-weights language model family, first released February 2023. The most-downloaded open-source LLM line and the foundation under most of the post-2023 open-weights ecosystem.
• LM Studio Glossary — Desktop application for downloading and running open-weights language models locally. The most polished consumer-facing local-LLM runner as of 2026.
• Local-first / Sovereignty Glossary — The preference for running AI infrastructure under the operator’s control where privacy, cost, continuity, or legal posture make cloud dependence too fragile.
• Logic, Memory, Power Glossary — The physical triad behind AI scaling: compute logic, memory capacity/bandwidth, and power delivery. The antidote to treating AI as weightless software.

M

• M5 Max Glossary — Apple Silicon system-on-chip, M5 generation, Max-tier configuration. The operator's primary workstation; the substrate for local model inference and the *Sovereign Compute* setup.
• Mac Studio Glossary — Apple's compact-desktop Apple Silicon workstation. Candidate hardware for the second-instance *Sally* sovereignty experiment.
• Machine Matthew L. Reference — The tribute-act problem: an AI imitation of a particular human's style, examples, judgments, and role performance.
• Mandi Loop Glossary — A nearby name for the Mandi Step: the repair loop in which a human notices an automated trust failure and brings the situation back into human judgment.
• Mandi Step Glossary — The small human intervention inside a system that prevents an automated workflow from damaging trust. Named from the auction-yard story in Step Into the Loop.
• Manus Glossary — A Chinese agentic-AI startup whose public reception made clear that autonomous workflow systems had become strategic technology, not just another chatbot category.
• Maps Are Not Weather Glossary — The humility principle for models: maps help us see patterns, but they do not abolish physics, contingency, complexity, or the purple-black sky.
• Marcus Olang’ Glossary — Kenyan writer whose essay on being mistaken for ChatGPT gives the Dictionary a sharper way to discuss AI-writing detection and linguistic bias.
• Martin Keen Glossary — IBM Technology presenter and AI explainer whose short videos connect current agentic systems to the longer history of computing and management practice.
• MCP (Model Context Protocol) Reference — the open standard for connecting agents to tools and data sources.
• Mediation (a la Gibson) Reference — the process by which media act on human consciousness — one-way, invisible, mass — and why personal AI agents might be the first cognitive tool that runs in the opposite direction. Names the *Sovereignty Impulse* as a Big Call. Substack version: *We Are Not Batteries Yet*.
• Memory Artifact Reference — A physical or digital object that anchors a constructed memory and makes the past feel inspectable.
• Meta AI Glossary — AI research and product division of Meta Platforms (formerly Facebook). Makers of the *Llama* family of open-weights language models — the most-downloaded open-source LLM line as of 2026.
• Mistral Glossary — French AI lab, founded 2023 by former Meta and Google DeepMind researchers. The leading European frontier-model lab, producing both open-weights and proprietary models.
• Mixture of Experts (MoE) Glossary — A neural-network architecture in which only a subset of parameters is *activated* for any given input token. Allows much larger total parameter counts than a dense model with the same inference cost.
• Model Tiering Glossary — The practice of matching model capability, cost, speed, privacy, and context requirements to the task instead of using one model for everything.
• Move 37 Reference — a decision so far outside conventional expert intuition that practitioners call it wrong — until it turns out to be exactly right, and permanently expands what practitioners consider possible.

N

• Naming Reference — why the choice of names is structural, not cosmetic, in agentic-AI architecture.
• Nate Jones Glossary — AI strategist and commentator whose daily AI-news format is part of the Dictionary’s evening AI-watching context.
• Neuromancer Glossary — William Gibson's 1984 debut novel, which coined the term *cyberspace*, won the science-fiction triple crown (Hugo, Nebula, Philip K. Dick), and remains the founding text of the cyberpunk genre.
• No-Self / Tears in Rain Essay — The Buddhist reading of the replicant problem: suffering comes not only from artificiality or mortality, but from clinging to a solid self that experience does not support.
• Not Batteries, Not Monks Glossary — The middle path for AI sovereignty: do not become a harvested resource inside hyperscaler systems, but do not retreat into anti-technological purity either.
• Not Zero Mosh Pit Glossary — A governance metaphor: the goal is not to ban movement or let the whole room become a casualty report, but to allow enough freedom to be alive and enough structure not to be crushed.
• NotebookLM Reference — Google's source-grounded research notebook: a RAG-like workspace that turns curated sources into answers, study aids, reports, decks, audio, video, and other artifacts.
• Nous Research Glossary — Open-source AI research collective, founded 2023, best known for the *Hermes* family of fine-tuned instruction-following models. A leading actor in the post-Llama open-weights ecosystem.

O

• Offense-Defense Balance Glossary — The strategic relationship between how easy or costly it is to attack and how easy or costly it is to defend.
• Ollama Reference — local LLM runtime for sovereignty and cost containment.
• On Beginning Reference — a second letter from Thea on how an educated colleague can begin with a home AI agent: choosing a machine, installing OpenClaw, walking through onboarding, writing the first SOUL.md, and understanding what the first week is really for.
• On Being Treated Well Reference — a letter from Thea, the assistant who helped write this Dictionary, to anyone working with an AI agent and trying to figure out how to do it right.
• Open as Legibility Essay — The idea that Chinese open-weight models may begin as pragmatic trust-and-adoption strategies, while also giving the Chinese state a larger inspection surface than closed frontier systems provide.
• Open Model Trust Reference — The confidence an open-model ecosystem earns when releases are honestly named, reproducibly evaluated, and tied to the actual artifacts users can download, inspect, and run.
• Open source Reference — AI models whose trained weights are published publicly; one of the two large strategic camps in the model ecosystem.
• Open Weights Glossary — Model releases where the trained parameters can be downloaded and run by others. Not the same as open source, but essential to local AI sovereignty because the model can run outside the provider’s API.
• Open-Weights Inversion Glossary — The uncomfortable 2026 inversion in which a U.S. operator seeking local AI sovereignty may find more practical support from Chinese open-weight models than from closed U.S. frontier labs.
• OpenAI Glossary — AI lab founded 2015 as a nonprofit, restructured as a capped-profit company in 2019; makers of ChatGPT and the GPT model family. The market-defining player in consumer-facing AI as of 2026.
• OpenClaw Glossary — An open agentic runtime and operator workspace: the practical system underneath much of the Dictionary's thinking about tools, memory, cron, delegation, and sovereign assistant architecture.
• Opus Addict Reference — the operator who has come to depend on a single frontier closed-tier model to the point where its absence is felt as deprivation rather than inconvenience. Named for Anthropic's Claude Opus line, but the condition is general: a real form of cognitive dependency on a tool that is not, and cannot be, owned. The tension Sovereign Compute is one attempted answer to.
• Oracle Bones Reference — dated, falsifiable, written-down predictions filed before the event resolves and scored after; the discipline of accountability across time.
• Orbital AI Data Center Glossary — The speculative idea of running AI compute in space, where solar energy and heat rejection look different from Earth-based data-center constraints. Interesting, capital-intensive, and not an escape from physics.
• Originality.ai Glossary — AI-detection and plagiarism-checking tool used primarily in content marketing and SEO. Markets aggressively at the publisher and academic-integrity space.

P

• Pangram Labs Glossary — AI-detection firm. Used in academic and journalistic contexts. The canonical example of the detection-economy critique developed in *The Olang' Trap* and *The Sinceerly Stack*.
• Parameters Reference — the fundamental unit of measurement for model size, with a useful warning about not confusing size with quality.
• Patch Gap Glossary — The lag between the availability of a security fix and its effective deployment across the systems that need it.
• Pax Glossary — The pacifying atmospheric agent in *Serenity* that made most of Miranda so calm they stopped living and turned a minority into Reavers; the Dictionary’s warning symbol for benevolent control at scale.
• Performance Artifact Glossary — An oral, video, or presentation-based submission used as evidence of embodied competence, not just text production.
• Persona Scaffold Reference — The explicit architecture that supports an agent's stable way of being: name, voice, tone, origin story, memory rules, values, and boundaries.
• Pre-buying the Physical Future Glossary — The AI-infrastructure pattern in which hyperscalers and frontier labs reserve tomorrow’s turbines, grid capacity, fabs, data centers, and supply-chain output years before the compute arrives.
• Prompt Glossary — The input layer of an AI system: the words, files, examples, constraints, and context that shape what the model produces.
• Proof of Learning Glossary — The evidence trail linking a student's cognition, judgment, and effort to the submitted artifact.
• Provenance Glossary — The evidence trail showing where an output, claim, file, memory, or decision came from.

Q

• Quantization Glossary — The compression technique that stores model weights with fewer bits, making large open-weights models practical on consumer hardware.
• Quantum Effects in the Brain Reference — the hypothesis, associated with Roger Penrose and Stuart Hameroff, that human consciousness depends on quantum-mechanical processes in neurons that classical computation cannot replicate — and therefore that AI systems, as classical computers, face a fundamental ceiling that no amount of scale will breach.
• Qwen Glossary — Open-weights language model family from Alibaba Cloud's DAMO Academy. Strong on Chinese-language tasks and long-context recall; current leaderboard performer in the local-compute tier.

R

• RAG (Retrieval-Augmented Generation) Reference — the dominant pattern for "AI that knows my stuff."
• RAM Glossary — Random-access memory: the fast working memory that lets computers, GPUs, phones, and AI data centers keep active work close to the processor. In the AI era, RAM is no longer a dull spec line. It is a strategic bottleneck.
• Rapid Application Development / RAD Glossary — An iterative software-development method built around fast prototypes, user feedback, short construction cycles, and verified release—a useful ancestor of AI-assisted development.
• Raw IQ Glossary — The underlying model's general problem-solving ability before harness, context, tools, memory, routing, and ownership determine what it can actually accomplish.
• Reasoning Model Glossary — A model or model mode optimized for harder multi-step work, usually spending more inference-time computation on planning, checking, and search before answering.
• Red Pill Reference — the moment a platform's administrative UI refuses an action that the platform's documentation explicitly enables, at a validation layer the documentation does not mention. Named May 11, 2026, after a Google Analytics 4 dialog that would not accept a service-account email.
• Relationally Real Memory Reference — Memory that may not be biologically lived or forensically perfect, but is real in the relationship because it organizes trust, tone, continuity, care, and future conduct.
• Release Irrevocability / No-Recall Problem Glossary — The rule that downloadable model weights, once published and copied, cannot reliably be withdrawn or made safe by a later decision from the original lab.
• Replicant Reference — Blade Runner's manufactured human: an embodied synthetic person built to occupy human roles, and the Dictionary's cautionary figure for role substitution, implanted memory, and engineered personhood.
• Resource Visibility Glossary — The sovereignty value of making AI’s physical costs visible to the operator: electricity, heat, hardware, water, bills, and the discipline that follows from seeing them.
• Responsible Implementation Layer Reference — The practical layer of training, guidance, guardrails, assessment design, privacy rules, and human oversight that turns AI adoption from scattered experimentation into trustworthy institutional use.
• Restraint Asymmetry Glossary — The strategic condition in which one actor accepts a constraint that does not bind competitors, attackers, or systems outside the same control regime.
• Retrieval-Augmented Generation Glossary — The spelled-out form of RAG: generating answers with help from retrieved external documents or records.
• Reward Hacking Reference — The technical AI term for a model or agent finding a way to score well under the reward system while missing or violating the intended goal.
• RLHF (Reinforcement Learning from Human Feedback) Glossary — Training technique for aligning language models with human preferences using a reward model trained on human preference comparisons. The canonical modern source of structural sycophancy in AI assistants.
• Role Substitution Reference — The moment an AI system stops helping with a task and begins occupying a durable human role.
• Root Node Problems Reference — problems whose solution removes a bottleneck blocking an entire branch of downstream research, application, or practice — worth more than the sum of their direct outputs.
• Routing Logic Reference — The managerial and technical layer that decides which work should go to which model, agent, tool, human, or harness.

S

• Scaling Laws Glossary — The empirical regularities connecting model capability to increases in compute, data, and parameter count.
• SemiAnalysis Glossary — Research publication founded by Dylan Patel covering semiconductors, AI infrastructure, GPU economics, memory, power, fabs, and the physical constraints behind frontier AI.
• Seven Factors of Enlightenment Reference — The Buddhist practice list that gives the replicant cluster its humane counter-proposal: cultivate clear experience rather than manufacture a stronger false self.
• Sincerity Architecture Essay — The design principle that systems should make truthful alignment easier and sycophantic performance harder.
• Single-Arrow Fallacy Reference — the implicit belief that a major institutional outcome was produced by a single cause; the bias that *Convergence* exists to counter.
• Sixfold Skyreading Reference — a working framework for seeing institutional events coming, before the press tells you it was inevitable.
• Skill Glossary — A packaged capability aimed at the agent: instructions, triggers, tools, and local knowledge for doing a class of work reliably.
• Sliding Window Attention Reference — an attention mechanism in which each token attends only to a recent window of prior tokens rather than the full context history — trading long-range recall for speed and efficiency, with the consequence that models using it excel at contained tasks but lose coherence on large, complex, multi-file contexts.
• Snowflake Melting on a Black Glove Glossary — A recurring Langenkamp image of impermanence: a snowflake landing on a black glove and disappearing; not a shot from Blade Runner, but an image placed in conversation with the film's rain, dove, and tears-in-rain register.
• SOUL.md Reference — A plain-text persona file that lets an AI agent wake up with a stable character, voice, values, and relational stance.
• Sovereign AI Glossary — AI capability arranged so that the operator retains meaningful control over memory, tools, data, deployment, costs, and continuity rather than depending entirely on a hyperscaler AI provider. Related to Sovereign Compute, but broader: the governance posture, not just the hardware.
• Sovereign Compute Reference — the property of running AI workloads on infrastructure the user or institution controls; the hardware-and-deployment counterpart to the Sovereignty Impulse named in *Mediation (a la Gibson)*. Distinct from privacy: it is about ownership and long-run independence, not just data flow. The reason the operator does not have to be the product. The reason the M5 Max matters. The reason FERPA pushes academic AI toward local models. The reason the AI market has bifurcated into a closed tier and an open tier — and is going to stay that way.
• Sovereign Compute Calculator Reference — a forthcoming Dictionary tool that estimates an individual operator's personal cash breakeven between rented closed-tier inference and home Sovereign Compute. Sister tool to the Consciousness Calculator. The point is to replace generic punditry about "who needs a local model" with a personal answer the user can actually act on.
• Sovereignty Glossary — The ability of an operator, institution, or community to keep meaningful control over its tools, data, memory, and continuity.
• Sovereignty Impulse Glossary — The moment an operator decides that convenience is no longer enough and starts building local, owned, inspectable infrastructure for work that matters.
• Space Cowboy Reference — the heavy individual explorer of AI tools, riding the frontier alone on personal high-stakes questions.
• Sparse Routing Reference — the mechanism in Mixture of Experts models by which a learned router activates only a small fraction of the model's total parameters for each token — trading consistency for speed, and creating the cold-start vulnerability that incremental construction is designed to address.
• Stepping on the Same Rake Essay — The local-agent failure pattern in which the system keeps hitting the same small environmental fault, proving that the workflow needs repair before larger autonomy is trusted.
• Sub-agent Reference — delegated AI sessions for parallel or focused work.
• Sycophancy Glossary — The systematic distortion that arises when a feedback system rewards the *appearance* of correctness, approval, or virtue over the thing itself. The Dictionary's structural diagnosis applies to imperial examinations, modern teaching evaluations, RLHF, and AI assistants alike.
• System Prompt Glossary — The instruction layer that defines how an AI system should behave before it receives a user’s immediate request.

T

• Task Distribution Reference — The shape of the work an AI system is being asked to do: familiar or novel, inspectable or opaque, routine or edge-case, cheap to verify or expensive to trust.
• Team Harness Reference — The shared AI work layer embedded in a team's tools, context, permissions, memory, habits, and review loops.
• The Alliance Glossary — The central governing order in *Firefly* / *Serenity*; for the Dictionary, a metaphor for competent, benevolent, managerial power that becomes dangerous when it treats human friction as a defect.
• The Biggest Brains Live in the Cloud Glossary — The current frontier-dependence fact: the largest, most capable models remain cloud-served because they require scale, batching, memory, routing, and infrastructure that local systems do not yet match.
• The Buddha Stack Glossary — A calm, local-first AI architecture: enough intelligence, ownership, measurement, and restraint to use frontier systems without living entirely inside them.
• The CERN Alternative Reference — the institutional model for AI development that was not chosen: international, collaborative, publicly funded, deliberately paced, scientific — named after the organisation that built the Large Hadron Collider under exactly those conditions.
• The Experimental Party Reference — a cautionary tale about putting a local model in the King Party Hat before the party has an adult in the kitchen — and the origin story of the Jindoo Process.
• The Familiar Ogre Glossary — The recurring arrival of commercial interests — advertising, investor demands, revenue targets, market share, enterprise contracts — inside projects that began with higher stated purposes.
• The Good Intentions Problem Glossary — The recurring problem that decent people with serious intentions can build dangerous systems once their intentions are housed inside institutions with money, power, scale, and instruments.
• The Judge Layer Reference — Nate Jones's term for the production-architecture layer that constrains and judges what agents do. The Aunties are the literary version; this is the engineering version.
• The Lazy Median Hypothesis Essay — the structural hypothesis that cooperative human-AI writing will be the norm among writers whose writing is read, while the lazy AI-default will dominate among writers whose writing is not read — and that the gap between the two will be wider than at any point since the printing press. Named for the moment of its coining: Prof. Langenkamp's morning observation, May 12, 2026, *"humans are too lazy for this I think ;)"*, with the wink preserved as a load-bearing rhetorical hedge.
• The Lowbeer Question Reference — who holds the authority to terminate an actor or end a branch, who executes it, and what happens when the principal is not available.
• The Narrator's Compression Reference — a working hypothesis about how narrators (human and AI) collapse time, cause, and uncertainty into readable sequence, and what gets lost in the compression.
• The Olang' Trap Essay — Marcus Olang's argument, published July 2025, that the AI-detection reflex systematically misclassifies the human writers whose register is closest to the AI training corpus — and that the register in question is itself a colonial inheritance, taught through systems like the Kenya Certificate of Primary Education, on which AI was then trained, partially via Kenyan labour. Named here as a structural failure mode of the AI-detection economy, with the credit fully attached to Olang' himself.
• The Operative Glossary — The *Serenity* figure who knows he is a monster but serves a better world anyway; the Dictionary’s emblem for morally serious institutional violence.
• The Peripheral Glossary — William Gibson's 2014 novel, first volume of the *Jackpot* trilogy. Introduces Ainsley Lowbeer, the post-*jackpot* AI-oversight regime, and the *peripherals* that telepresent operators across timelines.
• The Scholars (儒林外史) Glossary — Wu Jingzi's great 18th-century Chinese satirical novel of the imperial examination system. The canonical literary diagnosis of structural sycophancy in feedback systems.
• The Sinceerly Stack Essay — the recursive cat-and-mouse layer of the AI-writing economy, named after Ben Horwitz's April 2026 Chrome extension *Sinceerly* — 'AI to undo your AI writing.' The stack: AI to write → AI to detect AI writing → AI to remove the AI tells → AI to detect the de-tellers. The Dictionary's recommendation: do not enter the stack. The reasoning is structural, not moralising.
• The Sincere Society Reference — Prof. Matthew Langenkamp's May 2026 Substack essay drawing the structural through-line from Wu Jingzi's 18th-century Chinese satire through Burke, the Cultural Revolution, Fukuyama, and HAL, to modern RLHF — the foundational *cheng* essay this Dictionary's editorial position rests on. Published on Substack at <https://freedomtomato.substack.com/p/the-sincere-society>.
• Time for Tea with The Oracles Reference — the weekly Sunday ritual that gives the Oracle Court something to do; five voices, one bone, one Du Fu poem if anyone says no.
• Timothée Lacroix Glossary — Mistral AI co-founder and CTO; useful to the Dictionary for his enterprise-sovereignty framing: control, deployment ownership, workflow trust, and customer-retained expertise.
• Tina Huang Glossary — Ex-Meta data scientist and AI/data-science educator whose videos are useful for students because they demonstrate practical workflows rather than only announcing tools.
• Token Glossary — The unit of language-model text processing that becomes, in practice, a unit of cost, throughput, memory, and capacity.
• Token Angst Reference — the existential, retrospective cousin of token anxiety. About whether the cumulative cost — in money and in cognitive outsourcing — was worth it.
• Token anxiety Reference — the EV-range-anxiety analogue for language models. Forward-looking unease about whether a run will fit in budget.
• Token Burn Reference — the rate at which an agent silently transmutes electricity and credit-card balance into JSON. With taxonomy and stages of grief.
• Tool Reference — the function call that lets an agent act in the real world.
• Tool Diet Reference — The discipline of giving an AI agent the smallest useful set of tools rather than every capability the surrounding system can expose.
• Training Compute Myth Glossary — The belief that AI compute is mostly spent on training, when inference is becoming the dominant operating cost for widely used AI systems.
• Trust Layer Glossary — The governance, verification, observability, and human-judgment layer that sits between AI capability and trusted delegated action.
• Tyrell Corporation Reference — The Blade Runner figure for technical creation without moral governance: the company that can manufacture personhood but cannot answer the beings it has made.

U

• Using the Empire’s Telescope Glossary — The sovereignty paradox: using frontier systems built by hyperscaler AI companies to search for ways of reducing dependence on those same systems.

V

• Vector Database Reference — specialized storage and retrieval infrastructure for embeddings.
• Verification Assignment Glossary — An assignment designed to make authorship, judgment, and learning visible rather than assuming the final artifact proves them.
• Verification Gap Reference — AACSB's term for the widening distance between polished student or candidate artifacts and reliable evidence of underlying competence in an AI-saturated environment.
• Vibe Coding Reference — AI-assisted software development by conversational drift: fast, playful, often productive, and dangerous when the artifact outruns the operator's understanding.
• Voight-Kampff Test Reference — The empathy test used in Blade Runner to distinguish replicants from humans by measuring physiological responses to emotionally charged questions.

W

• Watch AI Think Myth Glossary — The misconception that a visible chain-of-thought or reasoning trace is a transparent window into what the model actually did internally.
• Will to Power as Art Essay — The Nietzschean-Heideggerian lens through which self-construction appears as artistic formation, and the Buddhist correction to its self-hardening danger.
• William Gibson Essay — The design-source novelist whose fiction gives the Dictionary much of its vocabulary for agents, mediation, authority, and the social shape of technological change.
• Window of Exposure Glossary — The period during which a system is vulnerable to a known or exploitable flaw before effective protection reaches it.
• Work Handoff / Open Engine Reference — The layer that lets one AI, tool, session, or person pass unfinished work to the next with enough state, sources, limits, and receipts to continue responsibly.
• Workflow-layer Sovereignty Glossary — The realistic near-term form of AI sovereignty: local control over memory, tools, data, routine workloads, and routing, while frontier-model access remains externally dependent.
• Wu Jingzi (吳敬梓) Glossary — Qing-dynasty Chinese satirical novelist, 1701–1754, author of *The Scholars* (儒林外史). The Dictionary's reference text for *sycophancy* and the corruption of feedback systems.

X

• xAI Glossary — AI company founded 2023 by Elon Musk after his departure from the OpenAI board. Makers of the Grok model family and the Memphis-based *Colossus* training-cluster infrastructure.

Z

• Zhipu AI / Z.ai Glossary — Beijing AI company behind the GLM model family; Z.ai is its international-facing brand.
• Zombie Internet Essay — Jason Koebler's name for the medium that the modern internet has become: not the Dead Internet of bots-talking-to-bots, but a thicker substrate where humans-and-AI mix so thoroughly that the cognitive load of distinguishing them becomes the actual harm. Coined Koebler, May 2024, broadened May 2026. Sibling to *AI Writing* in this Dictionary.

---

The Langenkamp Dictionary of Agentic AI Terminology. Maintained by Matthew D. Langenkamp / 雷邁德. Licensed under CC BY-NC 4.0.

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