{ "model": { "rank": 4806, "code": "thing-q115305900", "model_id": "vr.tr.large-language-model", "name": "large language model", "purpose": "Let an agent explain large language models and how they work in general terms, relay architectures, training and evaluation from research sources with attribution, describe capabilities, limitations and governance debates neutrally, and distinguish large language models from earlier language models, search engines and general artificial intelligence.", "family": "Thing Registry", "category": "Information and virtual systems", "status": "research-draft", "kind": "thing", "plane": "INF", "domain": "INF.KNW", "industry": "", "version": "", "url": "/models/thing/q115305900/", "tier": 2, "score": 72, "payload": { "layer": "wikidata", "aliases": [ "large action model", "GLM", "mixture of experts model", "generative pre-trained transformer", "Claude", "instruction model", "Gemini", "family of large language models", "multimodal large language model", "reasoning language model", "1.58-bit large language model", "Molmo", "DNA large language model", "custom GPT", "Claude Mythos Preview", "Claude Mythos" ], "aliasCount": 16, "merged": 16, "knownIn": 72, "facets": null, "markers": [], "lexicalClass": "", "senseRank": null, "alsoRegisteredAs": null, "source": { "dataset": "wikidata", "item": "Q115305900", "url": "https://www.wikidata.org/wiki/Q115305900", "license": "CC0 1.0" } }, "research": { "vercy": "1.0-draft", "publication": { "status": "research-draft", "adjudicationStatus": "unreviewed", "publishableCanonical": false, "generatedAt": "2026-09-13T05:23:21Z", "providers": [ "Claude" ], "breadth": "written by Claude from model knowledge without web access - no source was read, every claim is a lead to verify", "pass": 2, "wave": 3, "engine": "claude" }, "metaModel": { "id": "THING-Q115305900", "registryId": "vr.tr.large-language-model", "name": "large language model", "version": "0.2.0-wave.3", "entryKind": "thing", "family": "Thing Registry", "domain": [ "INF.KNW" ], "status": "research-draft" }, "canonicalUrl": "https://ver.cy/models/thing/q115305900/", "model": { "registry_id": "vr.tr.large-language-model", "name": "large language model", "purpose": "Let an agent explain large language models and how they work in general terms, relay architectures, training and evaluation from research sources with attribution, describe capabilities, limitations and governance debates neutrally, and distinguish large language models from earlier language models, search engines and general artificial intelligence.", "definition": "A type of language model built as a neural network with billions or more parameters, usually a transformer, trained on very large text corpora to predict tokens and then adapted through instruction tuning and feedback to follow prompts, in families such as generative pre-trained transformers and models from several developers including GPT, Claude, Gemini, Llama and GLM, with architectures including mixture of experts and extensions such as large action models that operate tools; large language models generate and analyse text, code and, in multimodal versions, images and audio, and raise questions of accuracy, bias, safety and governance.", "what_it_is_for": "Generating and processing language for many tasks.", "affordances": [ "explain how they work", "relay architectures and training", "describe capabilities and limits", "distinguish related systems" ], "distinguishing_features": [ "Transformer architecture", "Pretraining on large corpora", "Instruction tuning", "Emergent general capabilities" ], "appearance": "Not a visible object; software accessed through interfaces and APIs.", "visual_identification": [ "Neural language model with very many parameters", "GPT-style models, Claude, Gemini, Llama, GLM, mixture of experts models, instruction-tuned models, large action models", "Small language models are smaller; search engines retrieve rather than generate; general AI is a broader aspiration" ], "physical_properties": [ { "quantity": "parameters", "typical_range": "billions to trillions", "unit": "count", "note": "" }, { "quantity": "transformer introduced", "typical_range": "2017", "unit": "year", "note": "Attention Is All You Need" }, { "quantity": "context windows", "typical_range": "thousands to millions", "unit": "tokens", "note": "varies by model" } ], "families_and_kinds": [ "generative pre-trained transformers and other decoder models", "instruction and chat models", "mixture of experts models", "multimodal models", "large action models and agents that use tools", "open-weight and proprietary models from various developers" ], "related_models": [ { "relation": "is a kind of", "target": "language model", "why": "in registry terms" }, { "relation": "is built on", "target": "transformer (machine learning)", "why": "architecture" }, { "relation": "is trained by", "target": "reinforcement learning from human feedback", "why": "among other methods" }, { "relation": "is contrasted with", "target": "search engine", "why": "which retrieves documents" } ], "identifiers": [], "standards_and_regulation": [ "EU AI Act and other emerging AI regulations", "Model documentation practices such as model cards", "Copyright and data protection law applied to training data, which is contested" ], "failure_modes_and_hazards": [ "Hallucinated or inaccurate output", "Bias and harmful content", "Overreliance without verification", "Presenting any developer s claims as settled fact" ], "in_scope": [], "out_of_scope": [], "characteristics": [] }, "sources": [], "structure": { "bundles": [ { "id": "understand", "name": "Understand", "description": "What a large language model is.", "rationale": "Science.", "layers": [ { "id": "definition", "name": "Definition", "description": "Definition.", "findings": [ { "id": "definition-finding", "name": "Definition", "description": "Definition.", "questions": [ { "text": "What is a large language model, and how does it differ from earlier language models, search engines and general AI?", "kind": "definition" }, { "text": "Is the question about the technology in general, a specific product or a policy debate?", "kind": "boundary" } ] } ] }, { "id": "working", "name": "Working", "description": "How it works.", "findings": [ { "id": "working-finding", "name": "Working", "description": "Working.", "questions": [ { "text": "How do tokens, transformers, pretraining, instruction tuning and feedback produce a usable model, in general terms?", "kind": "definition" }, { "text": "Which entry fits the specific component?", "kind": "action" } ] } ] } ] }, { "id": "kinds", "name": "Kinds", "description": "Kinds and models.", "rationale": "Sources.", "layers": [ { "id": "architectures", "name": "Architectures", "description": "Architectures.", "findings": [ { "id": "architectures-finding", "name": "Architectures", "description": "Architectures.", "questions": [ { "text": "How do dense, mixture of experts, multimodal and action-oriented models differ?", "kind": "provenance" }, { "text": "Which references are standard?", "kind": "provenance" } ] } ] }, { "id": "models", "name": "Models", "description": "Notable models.", "findings": [ { "id": "models-finding", "name": "Models", "description": "Models.", "questions": [ { "text": "What are the major model families from different developers, described neutrally?", "kind": "provenance" }, { "text": "Is the presentation neutral across developers?", "kind": "boundary" } ] } ] } ] }, { "id": "use", "name": "Use", "description": "Use and limits.", "rationale": "Practice.", "layers": [ { "id": "capabilities", "name": "Capabilities", "description": "Capabilities.", "findings": [ { "id": "capabilities-finding", "name": "Capabilities", "description": "Capabilities.", "questions": [ { "text": "What can large language models do, and how are they evaluated?", "kind": "provenance" }, { "text": "Which sources are cited?", "kind": "provenance" } ] } ] }, { "id": "limits", "name": "Limits", "description": "Limitations.", "findings": [ { "id": "limits-finding", "name": "Limits", "description": "Limits.", "questions": [ { "text": "What are hallucination, bias, prompt injection and other limitations, and how are they mitigated?", "kind": "provenance" }, { "text": "Which entry fits AI safety?", "kind": "action" } ] } ] } ] }, { "id": "context", "name": "Context", "description": "Governance and history.", "rationale": "Context.", "layers": [ { "id": "governance", "name": "Governance", "description": "Governance.", "findings": [ { "id": "governance-finding", "name": "Governance", "description": "Governance.", "questions": [ { "text": "What debates surround regulation, copyright, energy use and labour impacts, with positions attributed?", "kind": "provenance" }, { "text": "Is the presentation neutral?", "kind": "boundary" } ] } ] }, { "id": "history", "name": "History", "description": "History.", "findings": [ { "id": "history-finding", "name": "History", "description": "History.", "questions": [ { "text": "How did language models develop from n-grams and RNNs to transformers and today s models?", "kind": "provenance" }, { "text": "Which entry fits the history of natural language processing?", "kind": "action" } ] } ] } ] } ] }, "openQuestions": [ "Should transformer and instruction tuning be separate primary entries?", "How should research sources be linked?", "The registry entry has merged aliases naming specific products and architectures; should they be split off?" ], "statistics": { "bundles": 4, "layers": 8, "findings": 8, "questions": 16 } }, "draft": { "generator": "vr.draft.v3", "status": "draft-generated", "researched": false, "archetype": "discipline or field of knowledge", "method": "Written from the archetype playbook - what this kind of thing needs beyond identity and provenance - and from the structure that recurred across 6,333 models already researched by two engines. Applied to this entry by rule. No source was read for this thing and no claim here is researched. This entry carries no facets of its own, so they were inferred from its domain - a guess about a whole domain applied to one thing.", "facetsInferred": true, "nextPass": "A researcher replaces this draft with a sourced specification. Treat every sentence below as a proposal to argue with.", "purpose": "Give an agent a durable, checkable way to recognise a large language model, record what state it is in, and decide what may be done with it.", "whatItIs": "Let an agent explain large language models and how they work in general terms, relay architectures, training and evaluation from research sources with attribution, describe capabilities, limitations and governance debates neutrally, and distinguish large language models from earlier language models, search engines and general artificial intelligence.", "characteristics": { "substance": "information", "origin": "conceptual", "agency": "inert" }, "whatYouCanDoWithIt": [ "read and interpreted" ], "distinguishingFeatures": [ "Names folded into this entry, which a task may need to split apart again: large action model, GLM, mixture of experts model, generative pre-trained transformer, Claude, instruction model, Gemini, family of large language models, multimodal large language model, reasoning language model, 1.58-bit large language model, Molmo.", "16 finer distinctions are held as aliases rather than separate entries, because telling them apart needs a task that asks for it.", "Described in 72 Wikipedia languages, which is a measure of how widely the thing is known, not of how important it is." ], "openQuestionsForResearch": [ "Which of the bundles below does a real task actually need, and which are ceremony?", "What does this thing have that the facets do not capture at all?", "Which neighbouring kind is most often confused with a large language model, and on what evidence are they told apart?" ], "whatItIsMadeOf": "content that has to be carried by something else", "physicalCharacter": [ "Does nothing on its own; everything it does, something else did to it.", "These come from the domain this entry sits in rather than from the entry itself, so treat them as a first guess about the whole domain applied to one thing." ], "whatCanBeDoneWithIt": [ "read it and act on what it says" ], "howItIsRecognised": [], "relatedModels": [ { "relation": "covers", "note": "Finer kinds folded into this entry because telling them apart needs a task that asks for it. Each is a model waiting to be split out when one does.", "targets": [ "large action model", "GLM", "mixture of experts model", "generative pre-trained transformer", "Claude", "instruction model", "Gemini", "family of large language models", "multimodal large language model", "reasoning language model", "1.58-bit large language model", "Molmo" ] } ], "standing": "Described in 72 Wikipedia languages, which measures how widely it is written about rather than how important or how common it is. 16 finer distinctions are held inside this entry as names rather than as separate models.", "structure": { "bundles": [ { "id": "identity-and-classification", "name": "Identity, naming and classification", "description": "How an agent tells one large language model from another, and a large language model from things that resemble it.", "rationale": "Recognition comes before every other claim. Without stable identity nothing else in the model can be trusted to be about the same thing twice.", "layers": [ { "id": "naming-and-identifiers", "name": "Names and identifiers", "description": "The names this thing goes by and the identifiers that survive translation and time.", "findings": [ { "id": "preferred-name-and-aliases", "name": "Preferred name, aliases and local names", "description": "Which name to use, which names mean the same thing, and which merely sound similar.", "questions": [ { "id": "preferred-name-and-aliases-q01", "text": "What identifies and describes the name of a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "preferred-name-and-aliases-q02", "text": "Who or what asserted this about the name of a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "preferred-name-and-aliases-q03", "text": "What may an agent decide or do once the name of a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] }, { "id": "stable-identifiers", "name": "Stable identifiers and external keys", "description": "Identifiers that keep pointing at this kind of thing across systems and languages.", "questions": [ { "id": "stable-identifiers-q01", "text": "What identifies and describes an identifier for a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "stable-identifiers-q02", "text": "Who or what asserted this about an identifier for a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "stable-identifiers-q03", "text": "What may an agent decide or do once an identifier for a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] }, { "id": "classification-and-granularity", "name": "Classification and granularity", "description": "Where a large language model sits among kinds, and how finely a task needs to cut it.", "findings": [ { "id": "kind-and-parents", "name": "Kind, parents and neighbouring kinds", "description": "The classes this thing belongs to and the ones it is next to.", "questions": [ { "id": "kind-and-parents-q01", "text": "What identifies and describes the kind of a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "kind-and-parents-q02", "text": "Who or what asserted this about the kind of a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "kind-and-parents-q03", "text": "What may an agent decide or do once the kind of a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] }, { "id": "distinguishing-features", "name": "Distinguishing features", "description": "What separates a large language model from the things most often confused with it.", "questions": [ { "id": "distinguishing-features-q01", "text": "What identifies and describes what distinguishes a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "distinguishing-features-q02", "text": "Who or what asserted this about what distinguishes a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "distinguishing-features-q03", "text": "What may an agent decide or do once what distinguishes a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] } ] }, { "id": "state-and-lifecycle", "name": "State and lifecycle", "description": "The states a large language model passes through and the events that move it between them.", "rationale": "Most decisions about a thing depend on what state it is in now, which is a claim with a time on it, not a property.", "layers": [ { "id": "lifecycle-stages", "name": "Lifecycle stages", "description": "From coming into existence to ceasing to be one of these.", "findings": [ { "id": "stages-and-transitions", "name": "Stages and transitions", "description": "The stages worth naming and what moves a large language model between them.", "questions": [ { "id": "stages-and-transitions-q01", "text": "What identifies and describes the lifecycle of a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "stages-and-transitions-q02", "text": "Who or what asserted this about the lifecycle of a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "stages-and-transitions-q03", "text": "What may an agent decide or do once the lifecycle of a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] }, { "id": "observations-and-status", "name": "Observations and current status", "description": "What is observed about a large language model, how often and by whom.", "findings": [ { "id": "observation-record", "name": "Observation record", "description": "How an observation of a large language model is recorded so that it can be superseded rather than overwritten.", "questions": [ { "id": "observation-record-q01", "text": "What identifies and describes an observation of a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "observation-record-q02", "text": "Who or what asserted this about an observation of a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "observation-record-q03", "text": "What may an agent decide or do once an observation of a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] } ] }, { "id": "content-and-access", "name": "Content, encoding and access", "description": "What a large language model says, how it is encoded and who may read it.", "rationale": "An informational thing carries content that can be copied, versioned and withheld, none of which its physical carrier explains.", "layers": [ { "id": "content-and-encoding", "name": "Content and encoding", "description": "The content itself, its format and its language.", "findings": [ { "id": "content-and-format", "name": "Content, format and language", "description": "What a large language model contains and in what form it is held.", "questions": [ { "id": "content-and-format-q01", "text": "What identifies and describes the content of a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "content-and-format-q02", "text": "Who or what asserted this about the content of a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "content-and-format-q03", "text": "What may an agent decide or do once the content of a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] }, { "id": "access-and-rights", "name": "Access, rights and retention", "description": "Who may read, copy or change it, and for how long it is kept.", "findings": [ { "id": "access-rules", "name": "Access rules and retention", "description": "The permissions attached to a large language model and the period it survives.", "questions": [ { "id": "access-rules-q01", "text": "What identifies and describes access to a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "access-rules-q02", "text": "Who or what asserted this about access to a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "access-rules-q03", "text": "What may an agent decide or do once access to a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] } ] }, { "id": "subject-and-method", "name": "Subject matter and method", "description": "What large language model studies and how it establishes anything.", "rationale": "A field is identified by its questions and its methods, and those are what distinguish it from the fields next to it.", "layers": [ { "id": "subject", "name": "Subject matter", "description": "The questions it takes as its own.", "findings": [ { "id": "subject-record", "name": "Questions and objects of study", "description": "What it is about, stated so a neighbouring field can be told apart.", "questions": [ { "id": "subject-record-q01", "text": "What questions and objects does large language model take as its own?", "kind": "definition" }, { "id": "subject-record-q02", "text": "Which questions does it share with a neighbouring field, and who claims them?", "kind": "boundary" } ] } ] }, { "id": "method", "name": "Method and evidence", "description": "How claims are established and what counts as evidence.", "findings": [ { "id": "method-record", "name": "Methods and standards of evidence", "description": "The methods used and what they are taken to establish.", "questions": [ { "id": "method-record-q01", "text": "By what methods does large language model establish claims, and what counts as sufficient evidence?", "kind": "provenance" }, { "id": "method-record-q02", "text": "What kind of claim can this field not settle, and where should an agent look instead?", "kind": "action" } ] } ] } ] }, { "id": "institutions-and-classification", "name": "Institutions and classification", "description": "Where large language model is practised and how catalogues place it.", "rationale": "The institutional footprint is the evidence that a field exists as a field rather than as a topic.", "layers": [ { "id": "institutions", "name": "Institutional markers", "description": "Societies, journals, degrees and departments.", "findings": [ { "id": "institution-record", "name": "Bodies and venues", "description": "Where the field organises itself.", "questions": [ { "id": "institution-record-q01", "text": "Which societies, journals or degrees mark large language model as an established field?", "kind": "provenance" }, { "id": "institution-record-q02", "text": "What would show that it is emerging or dissolving rather than established?", "kind": "boundary" } ] } ] }, { "id": "classification", "name": "Classification schemes", "description": "Where library and research classifications put it.", "findings": [ { "id": "classification-record", "name": "Codes and their disagreements", "description": "The classification codes that place it, and where they differ.", "questions": [ { "id": "classification-record-q01", "text": "Which classification schemes place large language model, under what codes?", "kind": "definition" }, { "id": "classification-record-q02", "text": "Where do those schemes disagree about its scope?", "kind": "boundary" } ] } ] } ] }, { "id": "provenance-and-evidence", "name": "Provenance, evidence and time", "description": "Where every claim about a large language model came from and when it held.", "rationale": "A claim without a source and a time cannot be superseded, only overwritten, and an agent that overwrites loses the ability to explain itself.", "layers": [ { "id": "source-and-authority", "name": "Source and authority", "description": "Who said it, on what evidence, and how strongly.", "findings": [ { "id": "claim-provenance", "name": "Claim provenance and confidence", "description": "The authority behind each claim about a large language model and how confident it is.", "questions": [ { "id": "claim-provenance-q01", "text": "What identifies and describes a claim about a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "claim-provenance-q02", "text": "Who or what asserted this about a claim about a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "claim-provenance-q03", "text": "What may an agent decide or do once a claim about a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] }, { "id": "time-and-versions", "name": "Time, versions and supersession", "description": "When a claim was true, when it was learnt, and what replaced it.", "findings": [ { "id": "validity-and-supersession", "name": "Validity period and supersession", "description": "How an old claim about a large language model is retired without being erased.", "questions": [ { "id": "validity-and-supersession-q01", "text": "What identifies and describes the validity of a claim about a large language model, and in what units or vocabulary?", "kind": "definition", "answer_data": [ "identifiers", "types and classes", "values with units", "explicit unknowns" ] }, { "id": "validity-and-supersession-q02", "text": "Who or what asserted this about the validity of a claim about a large language model, by which method, and when was it true?", "kind": "provenance", "answer_data": [ "authority", "method", "evidence", "event time", "knowledge time" ] }, { "id": "validity-and-supersession-q03", "text": "What may an agent decide or do once the validity of a claim about a large language model is known, and what must it refuse?", "kind": "action", "answer_data": [ "permitted actions", "preconditions", "refusals", "escalation" ] } ] } ] } ] } ] }, "statistics": { "bundles": 6, "layers": 12, "findings": 14, "questions": 38 } } } }