parametric knowledge
In plain wordsWhat an AI model 'knows' from training, stored in the numbers of its network rather than looked up in documents — like facts you have memorized rather than ones you check in a book.
Information encoded in a model's weightsweights, n.: The learned numbers inside a neural network that set how strongly each part influences the next. The finished set of weights is, in effect, the trained model itself. during training and available without any external source, as opposed to non-parametric or contextual knowledge supplied at inference time in the promptprompt, n.: The message or instructions you give an AI system — a question, a request, some examples or a document — that it responds to. Clearer prompts generally get better results. or by retrieval. Language models can answer many factual questions 'closed-book' from parametric knowledge, which can be probed with fill-in-the-blank queries; it is concentrated in, though not confined to, the feed-forward networkfeed-forward network, n.: The part of each transformer layer that works on every word separately, after attention has mixed in context. It holds most of a model's adjustable numbers and is thought to store much of its knowledge. layers, and can be located and altered by model editingmodel editing, n.: Changing one specific fact or behavior inside a trained AI model — for example, who leads a country — by directly adjusting a small part of it, without retraining the whole thing.. It is frozen at the knowledge cutoffknowledge cutoff, n.: The date after which an AI model knows nothing from its training, because its training data stops there. Without a web search tool, it cannot know about later events., unreliable for rare facts, hard to attribute to sources and costly to update. When parametric and contextual knowledge conflict, models may wrongly prefer their memorized answer; groundinggrounding, n.: Connecting an AI's words to something real — trusted documents, a live search, images or the physical world — so that its answers rest on evidence rather than only on what it absorbed in training. and retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from. are designed to let the context win.
Without retrieval it falls back on parametric knowledge, which is two years stale for this API.
Origin
Fabio Petroni and colleagues' 'Language Models as Knowledge Bases?' (2019) showed that pretrained models store retrievable facts. Patrick Lewis and colleagues' retrieval-augmented generation paper (2020) contrasted parametric with non-parametric memory, and Adam Roberts, Colin Raffel and Noam Shazeer (2020) measured how much knowledge fits in parameters; 'parametric knowledge' became standard from about 2021.
Related
- knowledge cutoffknowledge cutoff, n.: The date after which an AI model knows nothing from its training, because its training data stops there. Without a web search tool, it cannot know about later events.
- retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from.
- groundinggrounding, n.: Connecting an AI's words to something real — trusted documents, a live search, images or the physical world — so that its answers rest on evidence rather than only on what it absorbed in training.
- model editingmodel editing, n.: Changing one specific fact or behavior inside a trained AI model — for example, who leads a country — by directly adjusting a small part of it, without retraining the whole thing.
- memorizationmemorization, n.: When an AI model stores specific pieces of its training data — a paragraph, a phone number, an image — closely enough to reproduce them, instead of only learning general patterns.
- factualityfactuality, n.: How far what an AI says is actually true about the world. A factual model gets facts, numbers, names and dates right and does not make things up.
- reversal cursereversal curse, n.: A quirk of language models: having learned that 'A is B', they often cannot answer the reverse question — they may know who a celebrity's mother is yet fail to say who her son is.
- retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from.
- in-context learningin-context learning, n.: A language model's ability to pick up a new task from instructions or examples placed in its input, on the spot, without any retraining — like a quick study learning a game's rules from one round.
Sources
- Petroni, F. et al. (2019). Language Models as Knowledge Bases? EMNLP-IJCNLP 2019.
- Lewis, P. et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. NeurIPS 2020.
- Roberts, A., Raffel, C. & Shazeer, N. (2020). How Much Knowledge Can You Pack Into the Parameters of a Language Model? EMNLP 2020.
Mentioned in
agent memoryagent memory, n.: The ways an AI assistant or agent keeps information — your preferences, past conversations, lessons from earlier tasks — and brings it back later, beyond what fits in a single conversation., hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true., knowledge cutoffknowledge cutoff, n.: The date after which an AI model knows nothing from its training, because its training data stops there. Without a web search tool, it cannot know about later events., retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from., factualityfactuality, n.: How far what an AI says is actually true about the world. A factual model gets facts, numbers, names and dates right and does not make things up., groundinggrounding, n.: Connecting an AI's words to something real — trusted documents, a live search, images or the physical world — so that its answers rest on evidence rather than only on what it absorbed in training., long-term memorylong-term memory, n.: Memory that lasts. For an AI agent, it means information saved outside the conversation — facts, past events, learned procedures — that can be looked up and used again days or months later., model editingmodel editing, n.: Changing one specific fact or behavior inside a trained AI model — for example, who leads a country — by directly adjusting a small part of it, without retraining the whole thing., question answeringquestion answering, n.: Systems that reply to people's questions with the right answer, whether by finding it in a passage they are given, searching a database or the web, or drawing on what they already know., SimpleQASimpleQA, prop. n.: A quiz of about 4,300 brief factual questions, each with exactly one correct and stable answer, used to measure how often an AI gets facts right — and whether it admits when it does not know., tool usetool use, n.: An AI model's ability to reach beyond its own knowledge by operating outside tools — searching the web, running code, using a calculator or an app — and folding the results into its work., causal tracingcausal tracing, n.: A test that blurs the words of a factual question with noise and then restores one internal part of the model at a time, to find where in the model the fact gets recalled., episodic memoryepisodic memory, n.: Memory of particular past events — what happened, when and in what order. AI agents keep records of earlier conversations or attempts so they can recall them and learn from them., externalizationexternalization, n.: Moving an AI agent's know-how and memory out of the model itself and into files, tools and rules around it, so the whole system can do more than the model alone., knowledge neuronknowledge neuron, n.: An internal unit of a language model claimed to be responsible for storing a particular fact, such that turning it up or down makes the model more or less likely to state that fact., MEMITMEMIT, abbr.: A method for rewriting thousands of facts stored in a language model at once, by spreading small, calculated weight changes across several of its layers., reversal cursereversal curse, n.: A quirk of language models: having learned that 'A is B', they often cannot answer the reverse question — they may know who a celebrity's mother is yet fail to say who her son is., ROMEROME, abbr.: A method for rewriting a single fact stored in a language model — say, moving the Eiffel Tower to Rome — by making one small, precisely calculated change to one layer's weights.