# AI dictionary · Aula AIAO Reviewed on 5 November 2026 Original AIAO definitions and examples, checked against creators’ documentation, public bodies and research publications. Each entry links to a reference for further reading. TERMCAT’s basic vocabulary informs the Catalan names it covers. Other concepts use explanatory equivalents; this guide does not claim that every term is standardised. ## Accuracy ES: Exactitud · CA: Exactitud · EN: Accuracy The proportion of all predictions that are correct. It can hide poor results on rare categories. **An example makes it clearer** Getting 90 of 100 correct gives 90% accuracy; you still need to know which ten cases failed. Follow the connections: Precision, Recall Read more and verify: [scikit-learn · Accuracy score](https://scikit-learn.org/stable/modules/generated/sklearn.metrics.accuracy_score.html) ## AI agent ES: Agente de IA · CA: Agent d'IA · EN: AI agent A system using a model to decide steps and use tools to work towards a goal. **An example makes it clearer** It can find and compare courses; enrolling or paying requires separate permissions and controls. Follow the connections: Tool calling, Human-in-the-loop (HITL) Read more and verify: [Anthropic · Building effective agents](https://www.anthropic.com/research/building-effective-agents) ## AI alignment ES: Alineación de IA · CA: Alineació de la IA · EN: AI alignment Work to steer systems towards intended goals, values and constraints, checking gaps between intention and behaviour. **An example makes it clearer** Completing a task is not enough: how an assistant does it and what effects it causes also matter. Follow the connections: Reinforcement learning (RL), Red teaming Read more and verify: [IBM · AI alignment](https://www.ibm.com/think/topics/ai-alignment) ## AI bias ES: Sesgo de IA · CA: Biaix d'IA · EN: AI bias A systematic skew in data, design or outputs that can produce unfair representations or treatment. **An example makes it clearer** If nearly every “expert” example shows the same kind of person, examine the view the system is learning. Follow the connections: Data quality, AI evaluation Read more and verify: [IBM · AI bias](https://www.ibm.com/think/topics/ai-bias) ## AI evaluation ES: Evaluación de IA · CA: Avaluació d'IA · EN: AI evaluation Systematic assessment of a system's quality, failures and suitability using defined cases, criteria and metrics. **An example makes it clearer** You test ten FAQs and review content, sources and tone; “I liked it” is not the only criterion. Follow the connections: Benchmark, Human-in-the-loop (HITL) Read more and verify: [Anthropic · Define success criteria and build evaluations](https://platform.claude.com/docs/en/test-and-evaluate/develop-tests) ## AI hallucination ES: Alucinación de IA · CA: Al·lucinació d'IA · EN: AI hallucination An answer presenting false, invented or unsupported information as valid, even when it sounds convincing. **An example makes it clearer** The assistant cites a course that never existed. Giving it a name and timetable does not make it real. Follow the connections: Grounding, AI evaluation Read more and verify: [IBM · AI hallucinations](https://www.ibm.com/think/topics/ai-hallucinations) ## AI literacy ES: Alfabetización en IA · CA: Alfabetització en IA · EN: AI literacy Knowledge and skills to understand, use and evaluate AI thoughtfully, considering its possibilities and risks. **An example makes it clearer** It is not about memorising 120 words: it is knowing what to ask, what to check and when to seek help. Follow the connections: AI evaluation, Human-in-the-loop (HITL) Read more and verify: [Comisión Europea · AI literacy: questions and answers](https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers) ## AI model ES: Modelo de IA · CA: Model d'IA · EN: AI model A mathematical system fitted to turn inputs into predictions or outputs. It is one part of an application. **An example makes it clearer** The assistant's website is the interface; the model is part of the engine producing its reply. Follow the connections: Inference, Model parameter Read more and verify: [IBM · AI inference](https://www.ibm.com/think/topics/ai-inference) ## AI workflow ES: Flujo de trabajo con IA · CA: Flux de treball amb IA · EN: AI workflow A sequence combining models, tools and rules. It can follow a fixed path without being an autonomous agent. **An example makes it clearer** First summarise, then translate, then request review: three defined steps, not unlimited freedom. Follow the connections: Orchestration, AI agent Read more and verify: [Anthropic · Building effective agents](https://www.anthropic.com/research/building-effective-agents) ## Algorithm ES: Algoritmo · CA: Algorisme · EN: Algorithm A defined sequence of steps for solving a task. It can exist without AI or machine learning. **An example makes it clearer** Sorting courses by date follows an algorithm; not every clever button has AI inside. Follow the connections: AI model, Training Read more and verify: [IBM · Machine learning](https://www.ibm.com/think/topics/machine-learning) ## Anonymisation ES: Anonimización · CA: Anonimització · EN: Anonymisation Transformation that prevents reasonably identifying people from the data. Removing only the name is not always enough. **An example makes it clearer** A unique combination of age, town and role can reveal someone's identity; review those details too. Follow the connections: Pseudonymisation, Personal data Read more and verify: [AEPD · ¿Los datos anonimizados son datos personales?](https://www.aepd.es/preguntas-frecuentes/0-conceptos-basicos/FAQ-0005-sobre-los-datos-anonimizados) ## API endpoint ES: Punto de acceso de una API · CA: Punt d'accés d'una API · EN: API endpoint A specific address where a service receives requests for a particular operation. **An example makes it clearer** An API may have one address for generating text and another for checking a task's status. Follow the connections: API key, Tool calling Read more and verify: [MDN Web Docs · API](https://developer.mozilla.org/en-US/docs/Glossary/API) ## API key ES: Clave de API · CA: Clau d'API · EN: API key A credential identifying or authorising service use. Protect it because it may permit access or incur usage charges. **An example makes it clearer** Do not paste a real key into a public webpage's code; keep it in the application's secure environment. Follow the connections: Application programming interface (API), Data privacy Read more and verify: [Anthropic · API keys and authentication](https://platform.claude.com/docs/en/get-api-key) ## Application programming interface (API) ES: Interfaz de programación (API) · CA: Interfície de programació (API) · EN: Application programming interface (API) A set of rules and functions through which one program requests services or data from another. **An example makes it clearer** A website queries an API for available courses without manually copying all the information. Follow the connections: API endpoint, API key Read more and verify: [MDN Web Docs · API](https://developer.mozilla.org/en-US/docs/Glossary/API) ## Artificial intelligence (AI) ES: Inteligencia artificial (IA) · CA: Intel·ligència artificial (IA) · EN: Artificial intelligence (AI) Field that builds systems able to perform tasks such as recognising patterns, interpreting language or making decisions. **An example makes it clearer** A filter that detects suspicious messages uses AI; it does not need consciousness to do so. Follow the connections: Machine learning (ML), Generative AI Read more and verify: [IBM · Artificial intelligence](https://www.ibm.com/think/topics/artificial-intelligence) ## Artificial neuron ES: Neurona artificial · CA: Neurona artificial · EN: Artificial neuron A computing unit that combines inputs and applies a transformation. It is a mathematical simplification, not a living cell. **An example makes it clearer** Think of a small operation inside a network, not a tiny brain sipping coffee. Follow the connections: Neural network, Model parameter Read more and verify: [IBM · Deep learning](https://www.ibm.com/think/topics/deep-learning) ## Assistant memory ES: Memoria del asistente · CA: Memòria de l'assistent · EN: Assistant memory Information retained between steps or conversations, depending on the application. It differs from the context window and retraining. **An example makes it clearer** An assistant may save your preference for short answers; check what it stores and how to delete it. Follow the connections: Context window, Data privacy Read more and verify: [LangChain · Memory overview](https://docs.langchain.com/oss/python/concepts/memory) ## Attention ES: Atención · CA: Atenció · EN: Attention A mechanism that weights the relevance of different input parts when computing a representation. It does not imply consciousness. **An example makes it clearer** In “the mug beside the book is broken”, relationships between words help interpret what is broken. Follow the connections: Transformer, Context window Read more and verify: [Vaswani et al. · Attention Is All You Need](https://arxiv.org/abs/1706.03762) ## Automatic speech recognition (ASR) ES: Reconocimiento automático del habla · CA: Reconeixement automàtic de la parla · EN: Automatic speech recognition (ASR) Converting recorded or live speech into text. This differs from identifying who is speaking. **An example makes it clearer** You transcribe a talk for subtitles and review the names the system understood “in its own way”. Follow the connections: Text-to-speech (TTS), Natural language processing (NLP) Read more and verify: [Hugging Face · Automatic speech recognition](https://huggingface.co/tasks/automatic-speech-recognition) ## Benchmark ES: Prueba de referencia · CA: Prova de referència · EN: Benchmark A set of tasks and measurement rules for comparing systems. A high score does not guarantee performance everywhere. **An example makes it clearer** A model wins on maths problems, but you need to check how it summarises your lessons in Catalan. Follow the connections: AI evaluation, Test data Read more and verify: [Anthropic · Define success criteria and build evaluations](https://platform.claude.com/docs/en/test-and-evaluate/develop-tests) ## Chunking ES: Fragmentación de documentos · CA: Fragmentació de documents · EN: Chunking Splitting documents into manageable pieces for retrieval or processing while preserving enough context. **An example makes it clearer** You split a manual by section, avoiding separating a condition from the exception immediately after it. Follow the connections: Information retrieval, Context window Read more and verify: [IBM · Retrieval augmented generation](https://www.ibm.com/think/topics/retrieval-augmented-generation) ## Cloud-hosted model ES: Modelo en la nube · CA: Model al núvol · EN: Cloud-hosted model A model running on remote infrastructure accessed through an application or API under its terms. **An example makes it clearer** Your laptop sends the request to a service; review access, data handling and cost before using real information. Follow the connections: Local model, Data privacy Read more and verify: [Amazon Web Services · Deploy models for inference](https://docs.aws.amazon.com/sagemaker/latest/dg/deploy-model.html) ## Computer vision ES: Visión artificial · CA: Visió artificial · EN: Computer vision Techniques that extract and interpret information from images or video, such as objects, positions or movement. **An example makes it clearer** A system counts chairs in a room; it may be wrong when chairs obscure one another. Follow the connections: Optical character recognition (OCR), Multimodal model Read more and verify: [IBM · Computer vision](https://www.ibm.com/think/topics/computer-vision) ## Connector ES: Conector · CA: Connector · EN: Connector An integration connecting an AI application to another service, limited by its features and granted permissions. **An example makes it clearer** Connecting a notes folder should not automatically open your personal folders. Follow the connections: Model Context Protocol (MCP), Data privacy Read more and verify: [Model Context Protocol · What is MCP?](https://modelcontextprotocol.io/docs/getting-started/intro) ## Content Credentials ES: Credenciales de contenido · CA: Credencials de contingut · EN: Content Credentials Provenance information linked through cryptographic mechanisms using standards such as C2PA. It does not certify a scene is true. **An example makes it clearer** A signed history can show an image was edited; missing credentials do not prove it is fake either. Follow the connections: Content provenance, Watermark Read more and verify: [C2PA · Content Credentials explainer](https://spec.c2pa.org/specifications/specifications/2.3/explainer/Explainer.html) ## Content provenance ES: Procedencia del contenido · CA: Procedència del contingut · EN: Content provenance Information about content's origin and history, including known transformations. It does not alone establish truthfulness. **An example makes it clearer** Knowing which tool edited an image helps explain its history, but does not prove the scene happened. Follow the connections: Content Credentials, Deepfake Read more and verify: [C2PA · Content Credentials explainer](https://spec.c2pa.org/specifications/specifications/2.3/explainer/Explainer.html) ## Context window ES: Ventana de contexto · CA: Finestra de context · EN: Context window Maximum information a model can handle in an interaction, usually measured in tokens. **An example makes it clearer** It is the available workbench; fitting a whole book does not guarantee every detail gets handled well. Follow the connections: Token, Assistant memory Read more and verify: [Google Cloud · Generative AI glossary](https://docs.cloud.google.com/docs/ai-ml/generative-ai/glossary) ## Copyright ES: Derechos de autor · CA: Drets d'autor · EN: Copyright Rights protecting works and governing certain uses. Protection and exceptions depend on the context and applicable law. **An example makes it clearer** Finding an illustration online does not grant permission to put it in a course; check who permits what. Follow the connections: Licence, Content provenance Read more and verify: [World Intellectual Property Organization · Copyright](https://www.wipo.int/en/web/copyright) ## Data annotation ES: Anotación de datos · CA: Anotació de dades · EN: Data annotation Process of adding information to data, such as labels, transcripts or marked regions in an image. **An example makes it clearer** Drawing a box around each bicycle annotates the images; it does not retouch their colours. Follow the connections: Label, Computer vision Read more and verify: [IBM · Data labeling](https://www.ibm.com/think/topics/data-labeling) ## Data augmentation ES: Aumento de datos · CA: Augment de dades · EN: Data augmentation Creating variants of existing examples to diversify training while preserving what matters for the task. **An example makes it clearer** You slightly change the brightness of mug photos; they remain mugs, not new kinds of crockery. Follow the connections: Synthetic data, Overfitting Read more and verify: [IBM · Data augmentation](https://www.ibm.com/think/topics/data-augmentation) ## Data poisoning ES: Envenenamiento de datos · CA: Enverinament de dades · EN: Data poisoning Deliberate manipulation of data used by a system to alter behaviour, introduce bias or create vulnerabilities. **An example makes it clearer** Someone adds false training examples so the classifier learns an incorrect association. Follow the connections: Training data, Prompt injection Read more and verify: [OWASP · Data and model poisoning](https://genai.owasp.org/llmrisk/llm042025-data-and-model-poisoning/) ## Data privacy ES: Privacidad de los datos · CA: Privadesa de les dades · EN: Data privacy Protecting personal information's use and exposure: what is collected, why, who accesses it and how long it is kept. **An example makes it clearer** To practise summarising, use a fictional document rather than another person's real record. Follow the connections: Personal data, Assistant memory Read more and verify: [IBM · Data privacy](https://www.ibm.com/think/topics/data-privacy) ## Data quality ES: Calidad de los datos · CA: Qualitat de les dades · EN: Data quality How suitable data is for its intended use: accurate, complete, consistent and sufficiently current. **An example makes it clearer** A list with duplicate courses and outdated dates does not improve simply by feeding it to AI. Follow the connections: Dataset, AI bias Read more and verify: [IBM · Data quality](https://www.ibm.com/think/topics/data-quality) ## Dataset ES: Conjunto de datos · CA: Conjunt de dades · EN: Dataset An organised collection of examples used to analyse, train or evaluate a system. **An example makes it clearer** A folder of photos with a table of descriptions can form a dataset. Follow the connections: Label, Data quality Read more and verify: [Gebru et al. · Datasheets for Datasets](https://arxiv.org/abs/1803.09010) ## Dataset datasheet ES: Ficha del conjunto de datos · CA: Fitxa del conjunt de dades · EN: Dataset datasheet Documentation about a dataset's creation, composition, collection, uses and limitations. **An example makes it clearer** The datasheet explains where recordings came from, how they were selected and what they can be used for. Follow the connections: Dataset, Content provenance Read more and verify: [Gebru et al. · Datasheets for Datasets](https://arxiv.org/abs/1803.09010) ## Deep learning ES: Aprendizaje profundo · CA: Aprenentatge profund · EN: Deep learning Machine learning with multilayer neural networks that build increasingly complex representations. **An example makes it clearer** A network can combine edges, shapes and other patterns to recognise a bicycle in a photo. Follow the connections: Neural network, Computer vision Read more and verify: [IBM · Deep learning](https://www.ibm.com/think/topics/deep-learning) ## Deepfake ES: Suplantación audiovisual sintética · CA: Suplantació audiovisual sintètica · EN: Deepfake Image, video or audio manipulated or generated to convincingly resemble a real person or situation. **An example makes it clearer** A voice sounding like your friend asks for money: verify through another channel before believing it. Follow the connections: Content provenance, Content Credentials Read more and verify: [European Data Protection Supervisor · TechSonar: Deepfake detection](https://www.edps.europa.eu/system/files/2023-12/23-12-04_techsonar_23-24_en.pdf) ## Diffusion model ES: Modelo de difusión · CA: Model de difusió · EN: Diffusion model A generative model that learns to reverse a noise process to produce data such as images. **An example makes it clearer** During generation, an image can emerge from noise; this is not focusing a hidden photograph. Follow the connections: Text-to-image generation, Inpainting Read more and verify: [IBM · Diffusion models](https://www.ibm.com/think/topics/diffusion-models) ## Embedding ES: Representación vectorial · CA: Representació vectorial · EN: Embedding A numerical representation of data that allows comparison of learned relationships, such as similarities between texts. **An example makes it clearer** “I want to start with AI” can be close to “beginner course” without using identical words. Follow the connections: Semantic search, Vector database Read more and verify: [IBM · Vector databases](https://www.ibm.com/think/topics/vector-database) ## Epoch ES: Época de entrenamiento · CA: Època d'entrenament · EN: Epoch One complete pass through the training dataset. More epochs do not guarantee better results on new data. **An example makes it clearer** Three epochs go through the examples three times; unchecked repetition can also lead to overfitting. Follow the connections: Training batch, Overfitting Read more and verify: [Hugging Face · Fine-tuning](https://huggingface.co/docs/transformers/en/training) ## Explainability ES: Explicabilidad · CA: Explicabilitat · EN: Explainability Ability to provide understandable information about how or why a system produces outputs. A generated explanation may be inaccurate. **An example makes it clearer** You ask which factors influenced a classification and check the explanation with analysis methods, not just its narrative. Follow the connections: AI evaluation, Human-in-the-loop (HITL) Read more and verify: [IBM · Explainable AI](https://www.ibm.com/think/topics/explainable-ai) ## F1 score ES: Puntuación F1 · CA: Puntuació F1 · EN: F1 score The harmonic mean of precision and recall. It summarises their balance but does not replace examining specific errors. **An example makes it clearer** If you detect almost everything but raise many false alarms, F1 reflects some of that imbalance. Follow the connections: Precision, Recall Read more and verify: [scikit-learn · Classification metrics](https://scikit-learn.org/stable/modules/model_evaluation.html#precision-recall-f-measure-metrics) ## Feature ES: Característica de entrada · CA: Característica d'entrada · EN: Feature A property a model uses for prediction, measured directly or extracted from other data. **An example makes it clearer** When estimating a lesson's duration, the number of exercises could be a feature. Follow the connections: Label, Dataset Read more and verify: [IBM · Supervised learning](https://www.ibm.com/think/topics/supervised-learning) ## Few-shot prompting ES: Petición con pocos ejemplos · CA: Petició amb pocs exemples · EN: Few-shot prompting A request containing a few examples of the desired format or behaviour, without retraining the model. **An example makes it clearer** You add two messages and their classifications before the new message you want classified. Follow the connections: Zero-shot prompting, Fine-tuning Read more and verify: [Anthropic · Prompting best practices](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) ## Fine-tuning ES: Ajuste fino · CA: Afinament · EN: Fine-tuning Additional training that adapts an existing model to a task or domain. Unlike a prompt, it changes parameters. **An example makes it clearer** You train on reviewed examples of your organisation's style; adding a document to chat is different. Follow the connections: Prompt, Retrieval-augmented generation (RAG) Read more and verify: [Hugging Face · Fine-tuning](https://huggingface.co/docs/transformers/en/training) ## Foundation model ES: Modelo fundacional · CA: Model fundacional · EN: Foundation model A model trained on broad data that can be adapted to different tasks. Supported formats vary. **An example makes it clearer** One model can provide a foundation for summarising text, answering questions or classifying messages. Follow the connections: Large language model (LLM), Transfer learning Read more and verify: [IBM · Large language models](https://www.ibm.com/think/topics/large-language-models) ## Generative AI ES: IA generativa · CA: IA generativa · EN: Generative AI AI that produces text, images, audio or other content from learned patterns and supplied inputs. **An example makes it clearer** You ask for three workshop titles and get new suggestions; then you choose and review them. Follow the connections: Prompt, AI hallucination Read more and verify: [IBM · Artificial intelligence](https://www.ibm.com/think/topics/artificial-intelligence) ## Grounding ES: Fundamentación en fuentes · CA: Fonamentació en fonts · EN: Grounding Connecting an answer to verifiable reference information to reduce unsupported claims. **An example makes it clearer** You require each time to be backed by the published programme. A decorative citation is not enough. Follow the connections: Retrieval-augmented generation (RAG), AI hallucination Read more and verify: [Google Cloud · Generative AI glossary](https://docs.cloud.google.com/docs/ai-ml/generative-ai/glossary) ## Guardrails ES: Salvaguardas · CA: Salvaguardes · EN: Guardrails Measures limiting behaviour and risks: permissions, validation, filters, oversight and other complementary controls. **An example makes it clearer** Before sending anything, the application requires confirmation and checks the recipient and content. Follow the connections: Human-in-the-loop (HITL), Tool calling Read more and verify: [IBM · AI guardrails](https://www.ibm.com/think/topics/ai-guardrails) ## Human-in-the-loop (HITL) ES: Supervisión humana · CA: Supervisió humana · EN: Human-in-the-loop (HITL) A person's involvement in reviewing, deciding or authorising an automated process. **An example makes it clearer** AI drafts the email and you review it before sending. That last click belongs to someone. Follow the connections: AI agent, Guardrails Read more and verify: [IBM · Human-in-the-loop](https://www.ibm.com/think/topics/human-in-the-loop) ## Hyperparameter ES: Hiperparámetro · CA: Hiperparàmetre · EN: Hyperparameter A setting that controls training or architecture and is not learned like an ordinary weight. **An example makes it clearer** The learning rate controls adjustment size: changing it can improve or worsen training. Follow the connections: Model parameter, Epoch Read more and verify: [IBM · Hyperparameter tuning](https://www.ibm.com/think/topics/hyperparameter-tuning) ## Inference ES: Inferencia · CA: Inferència · EN: Inference Using an already trained model to produce an output from an input. **An example makes it clearer** Each time the model summarises your text it performs inference, without necessarily learning from that conversation. Follow the connections: Training, Latency Read more and verify: [IBM · AI inference](https://www.ibm.com/think/topics/ai-inference) ## Information retrieval ES: Recuperación de información · CA: Recuperació d'informació · EN: Information retrieval Finding documents or passages relevant to a query. Retrieving something does not make it correct. **An example makes it clearer** The search engine finds an old entry; check its date before using it as a current answer. Follow the connections: Reranking, Data quality Read more and verify: [IBM · Retrieval augmented generation](https://www.ibm.com/think/topics/retrieval-augmented-generation) ## Inpainting ES: Relleno generativo de una zona · CA: Emplenament generatiu d'una zona · EN: Inpainting Generating or replacing content within a selected image region, usually specified by a mask. **An example makes it clearer** You select a mug in a photo to replace it with a book, then check shadows and edges. Follow the connections: Outpainting, Text-to-image generation Read more and verify: [Hugging Face · Inpainting](https://huggingface.co/docs/diffusers/en/using-diffusers/inpaint) ## Jailbreak ES: Evasión de salvaguardas · CA: Evasió de salvaguardes · EN: Jailbreak An attempt to make a model ignore its safety or behavioural boundaries. **An example makes it clearer** Framing a prohibited request as a game does not remove the risk of what is being requested. Follow the connections: Prompt injection, Red teaming Read more and verify: [OWASP · LLM prompt injection prevention](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) ## Knowledge base ES: Base de conocimiento · CA: Base de coneixement · EN: Knowledge base An organised collection of information for reference or reuse. It can support an assistant without retraining its model. **An example makes it clearer** You gather Aula guides, FAQs and current programmes in one maintained, reviewed place. Follow the connections: Retrieval-augmented generation (RAG), Data quality Read more and verify: [Amazon Web Services · Knowledge bases for Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base.html) ## Knowledge distillation ES: Destilación de conocimiento · CA: Destil·lació de coneixement · EN: Knowledge distillation Training a model to imitate another model's behaviour, often to obtain a smaller version. **An example makes it clearer** A compact model learns from a larger model's answers; you must check which capabilities it retains. Follow the connections: Quantization, AI evaluation Read more and verify: [IBM · Knowledge distillation](https://www.ibm.com/think/topics/knowledge-distillation) ## Knowledge graph ES: Grafo de conocimiento · CA: Graf de coneixement · EN: Knowledge graph A representation of entities and their relationships for exploring explicit connections between concepts or data. **An example makes it clearer** “Course → contains → lesson” and “lesson → explains → RAG” form queryable connections. Follow the connections: Knowledge base, Information retrieval Read more and verify: [IBM · Knowledge graph](https://www.ibm.com/think/topics/knowledge-graph) ## Label ES: Etiqueta · CA: Etiqueta · EN: Label An answer or category assigned to an example and used as a reference in supervised learning. **An example makes it clearer** In a photo of a mug, “mug” can be the correct label. Follow the connections: Data annotation, Supervised learning Read more and verify: [IBM · Data labeling](https://www.ibm.com/think/topics/data-labeling) ## Large language model (LLM) ES: Gran modelo de lenguaje (LLM) · CA: Model de llenguatge gran (MLG) · EN: Large language model (LLM) A large-scale model trained to process and generate language. A fluent answer does not prove it is true. **An example makes it clearer** It helps draft a course description; you check that it does not invent the syllabus. Follow the connections: AI hallucination, Token Read more and verify: [IBM · Large language models](https://www.ibm.com/think/topics/large-language-models) ## Latency ES: Latencia · CA: Latència · EN: Latency The time an operation takes to respond. For generation, distinguish the start of the response from completion. **An example makes it clearer** The first words appear quickly but the complete text takes longer: these are different measurements. Follow the connections: Streaming response, Throughput Read more and verify: [IBM · AI inference](https://www.ibm.com/think/topics/ai-inference) ## Licence ES: Licencia de uso · CA: Llicència d'ús · EN: Licence Terms granting specific permissions over a resource. They may require attribution or limit modification and commercial use. **An example makes it clearer** Check the model licence, data licence and the licence of images you publish separately. Follow the connections: Open-source AI, Copyright Read more and verify: [Creative Commons · CC licences](https://creativecommons.org/cc-licenses/) ## Local model ES: Modelo local · CA: Model local · EN: Local model A model running on your own device. It needs sufficient resources and does not alone guarantee the whole application's privacy. **An example makes it clearer** Computation happens on your computer, but check whether the application sends logs or uses external services. Follow the connections: Quantization, Cloud-hosted model Read more and verify: [Ollama · Frequently asked questions](https://docs.ollama.com/faq) ## Machine learning (ML) ES: Aprendizaje automático · CA: Aprenentatge automàtic · EN: Machine learning (ML) Techniques that fit a model using data so it can perform tasks on new cases. **An example makes it clearer** Instead of writing every spam rule, you train on emails that have already been classified. Follow the connections: Training, Supervised learning Read more and verify: [IBM · Machine learning](https://www.ibm.com/think/topics/machine-learning) ## Mixture of experts (MoE) ES: Mezcla de expertos (MoE) · CA: Barreja d'experts (MoE) · EN: Mixture of experts (MoE) An architecture with multiple specialised blocks and a mechanism that selects which participate in each computation. **An example makes it clearer** These are not people advising the model: they are mathematical components activated selectively. Follow the connections: Model parameter, Transformer Read more and verify: [Hugging Face · Mixture of Experts Explained](https://huggingface.co/blog/moe) ## MLOps ES: Operaciones de aprendizaje automático · CA: Operacions d'aprenentatge automàtic · EN: MLOps Practices for developing, deploying, monitoring and updating models in a controlled, reproducible way. **An example makes it clearer** You track versions, check changes and monitor failures after release; the job does not end at “uploaded”. Follow the connections: AI evaluation, Model card Read more and verify: [Google Cloud · MLOps and delivery pipelines](https://docs.cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning) ## Model card ES: Ficha del modelo · CA: Fitxa del model · EN: Model card A document describing a model, intended uses, evaluation, limitations and other relevant information. **An example makes it clearer** Before choosing a model for Catalan, check which languages were evaluated and their results. Follow the connections: AI evaluation, Licence Read more and verify: [Hugging Face · Model cards](https://huggingface.co/docs/hub/en/model-cards) ## Model Context Protocol (MCP) ES: Protocolo de contexto de modelos (MCP) · CA: Protocol de context de models (MCP) · EN: Model Context Protocol (MCP) A standard for connecting AI applications to tools and data sources through a common interface. **An example makes it clearer** A connector can offer file search; the protocol does not itself grant permission to read every file. Follow the connections: Connector, Tool calling Read more and verify: [Model Context Protocol · What is MCP?](https://modelcontextprotocol.io/docs/getting-started/intro) ## Model parameter ES: Parámetro del modelo · CA: Paràmetre del model · EN: Model parameter An internal value learned during training, such as a weight. It is different from a written instruction. **An example makes it clearer** Changing “reply more briefly” changes the request; it does not directly rewrite model weights. Follow the connections: Hyperparameter, Fine-tuning Read more and verify: [IBM · Hyperparameter tuning](https://www.ibm.com/think/topics/hyperparameter-tuning) ## Multimodal model ES: Modelo multimodal · CA: Model multimodal · EN: Multimodal model A model that combines or processes more than one kind of information, such as text, images or audio. **An example makes it clearer** You upload a whiteboard photo and ask which tasks appear. Reading photos does not imply image-generation capability. Follow the connections: Computer vision, Automatic speech recognition (ASR) Read more and verify: [IBM · Large language models](https://www.ibm.com/think/topics/large-language-models) ## Natural language processing (NLP) ES: Procesamiento del lenguaje natural (PLN) · CA: Processament del llenguatge natural (PLN) · EN: Natural language processing (NLP) The field developing methods for computers to analyse, interpret or generate human language. **An example makes it clearer** Classifying enquiries, extracting names and translating are language tasks; not all need a large chatbot. Follow the connections: Large language model (LLM), Automatic speech recognition (ASR) Read more and verify: [IBM · Natural language processing](https://www.ibm.com/think/topics/natural-language-processing) ## Neural network ES: Red neuronal · CA: Xarxa neuronal · EN: Neural network A model of connected units that transform information through mathematical operations and learned parameters. **An example makes it clearer** A network distinguishes photos of mugs and plates by adjusting its connections during training. Follow the connections: Artificial neuron, Deep learning Read more and verify: [IBM · Deep learning](https://www.ibm.com/think/topics/deep-learning) ## Open weights ES: Pesos abiertos · CA: Pesos oberts · EN: Open weights A model whose trained parameters are distributed for download. Usage permissions depend on its licence and may include restrictions. **An example makes it clearer** You may have the weights without having the data or complete training process. Follow the connections: Open-source AI, Model parameter Read more and verify: [Open Source Initiative · The Open Source AI Definition 1.0](https://opensource.org/ai/open-source-ai-definition) ## Open-source AI ES: IA de código abierto · CA: IA de codi obert · EN: Open-source AI AI distributed with freedoms to use, study, modify and redistribute it, plus the necessary components, under the applicable definition. **An example makes it clearer** Being able to download a model is not enough to call it open source: check components and licensing. Follow the connections: Open weights, Licence Read more and verify: [Open Source Initiative · The Open Source AI Definition 1.0](https://opensource.org/ai/open-source-ai-definition) ## Optical character recognition (OCR) ES: Reconocimiento óptico de caracteres (OCR) · CA: Reconeixement òptic de caràcters (OCR) · EN: Optical character recognition (OCR) Converting text in images into machine-readable text. Recognition can contain errors and needs checking. **An example makes it clearer** You photograph a poster and extract its words; check dates, numbers and names especially carefully. Follow the connections: Computer vision, Data quality Read more and verify: [IBM · Optical character recognition](https://www.ibm.com/think/topics/optical-character-recognition) ## Orchestration ES: Orquestación · CA: Orquestració · EN: Orchestration Coordinating order, data exchange and failure handling among an AI system's components. **An example makes it clearer** If search fails, the system decides to retry or ask for help instead of inventing results. Follow the connections: AI workflow, Tool calling Read more and verify: [Anthropic · Building effective agents](https://www.anthropic.com/research/building-effective-agents) ## Outpainting ES: Ampliación generativa · CA: Ampliació generativa · EN: Outpainting Generating content beyond an image's original boundaries to extend its framing. **An example makes it clearer** You turn a square photo into a wide header by inventing its sides; those sides do not document the original scene. Follow the connections: Inpainting, Content provenance Read more and verify: [Hugging Face · Outpainting](https://huggingface.co/docs/diffusers/en/advanced_inference/outpaint) ## Overfitting ES: Sobreajuste · CA: Sobreajustament · EN: Overfitting The model fits training examples too closely and performs worse when it encounters new cases. **An example makes it clearer** It memorises the exercises, but change the numbers and it stares at the ceiling. Follow the connections: Test data, Underfitting Read more and verify: [IBM · Overfitting](https://www.ibm.com/think/topics/overfitting) ## Personal data ES: Datos personales · CA: Dades personals · EN: Personal data Information that identifies or can identify a person, alone or combined with other data. **An example makes it clearer** A name, address or combination of details may identify someone even without an identity number. Follow the connections: Data privacy, Anonymisation Read more and verify: [AEPD · ¿Qué son los datos personales?](https://www.aepd.es/preguntas-frecuentes/2-tus-obligaciones-como-responsable-del-tratamiento/2-aplicacion-de-la-normativa/FAQ-0241-que-son-los-datos-personales) ## Planning ES: Planificación · CA: Planificació · EN: Planning Breaking a goal into steps that can be executed and reviewed. Having a plan does not prove completion. **An example makes it clearer** “Review, fix and test the website” is the plan; seeing it work is another matter. Follow the connections: AI agent, AI evaluation Read more and verify: [Google Cloud · What are AI agents?](https://cloud.google.com/discover/what-are-ai-agents) ## Precision ES: Precisión · CA: Precisió · EN: Precision Among cases predicted positive, the proportion that truly is positive. It differs from overall accuracy. **An example makes it clearer** It flags ten messages as spam and eight really are: spam precision is 80%. Follow the connections: Recall, F1 score Read more and verify: [scikit-learn · Classification metrics](https://scikit-learn.org/stable/modules/model_evaluation.html#precision-recall-f-measure-metrics) ## Prompt ES: Prompt o instrucción · CA: Indicació o petició · EN: Prompt The input used to request a task from a model. It can include instructions, context, examples and other formats. **An example makes it clearer** “Summarise this announcement in three sentences for beginners and keep the date” is a specific request. Follow the connections: Prompt engineering, Few-shot prompting Read more and verify: [Anthropic · Prompt engineering overview](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview) ## Prompt caching ES: Caché de contexto · CA: Memòria cau de context · EN: Prompt caching Reusing processing for repeated parts of an input to reduce work, depending on the service's conditions. **An example makes it clearer** Several queries share the same lengthy instructions; caching can avoid recomputing them without becoming personal memory. Follow the connections: Context window, Assistant memory Read more and verify: [Anthropic · Prompt caching](https://platform.claude.com/docs/en/build-with-claude/prompt-caching) ## Prompt engineering ES: Ingeniería de prompts · CA: Enginyeria d'indicacions · EN: Prompt engineering Designing and revising requests through testing to obtain more useful, verifiable results. **An example makes it clearer** You replace “make it nice” with audience, format and criteria. Less telepathy, more context. Follow the connections: Prompt, Zero-shot prompting Read more and verify: [Anthropic · Prompt engineering overview](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview) ## Prompt injection ES: Inyección de instrucciones · CA: Injecció d'instruccions · EN: Prompt injection An attack inserting malicious instructions into inputs or retrieved content to redirect an AI system's behaviour. **An example makes it clearer** A retrieved document tries to give the assistant orders. The document is information, not the boss. Follow the connections: System prompt, Guardrails Read more and verify: [OWASP · Prompt injection](https://genai.owasp.org/glossary/prompt-injection/) ## Pseudonymisation ES: Seudonimización · CA: Pseudonimització · EN: Pseudonymisation Replacing identifiers to separate data from identity while retaining additional information that allows them to be linked again. **An example makes it clearer** You replace “Anna” with “Student 047” and keep the mapping separately. The data remains personal. Follow the connections: Anonymisation, Data privacy Read more and verify: [Comité Europeo de Protección de Datos · Anonimización y seudonimización](https://www.edpb.europa.eu/topics/ai-and-technology/anonymisation-pseudonymisation_es) ## Quantization ES: Cuantización · CA: Quantització · EN: Quantization Representing model values with lower numerical precision to reduce resource needs, potentially changing quality. **An example makes it clearer** A low-bit version may fit on your computer; test its results before assuming equivalence. Follow the connections: Local model, Knowledge distillation Read more and verify: [Hugging Face · Quantization](https://huggingface.co/docs/transformers/en/main_classes/quantization) ## Rate limit ES: Límite de solicitudes · CA: Límit de sol·licituds · EN: Rate limit A restriction on requests or processed data over a period. It differs from the available balance or budget. **An example makes it clearer** Even with available credit, too many requests in succession can require waiting and retrying in a controlled way. Follow the connections: Application programming interface (API), Throughput Read more and verify: [Anthropic · Rate limits](https://platform.claude.com/docs/en/api/rate-limits) ## Recall ES: Exhaustividad o sensibilidad · CA: Exhaustivitat o sensibilitat · EN: Recall The proportion of actual positives the system successfully detects for the category being evaluated. **An example makes it clearer** There are twenty spam messages and it detects fifteen: recall is 75%. Follow the connections: Precision, F1 score Read more and verify: [scikit-learn · Classification metrics](https://scikit-learn.org/stable/modules/model_evaluation.html#precision-recall-f-measure-metrics) ## Red teaming ES: Pruebas adversarias · CA: Proves adversàries · EN: Red teaming Deliberate evaluation seeking failures, misuse or dangerous behaviour within an authorised scope. **An example makes it clearer** A team tests difficult situations in a copy of the assistant before opening it to the public. Follow the connections: Robustness, Guardrails Read more and verify: [Anthropic · Challenges in red teaming AI systems](https://www.anthropic.com/news/challenges-in-red-teaming-ai-systems) ## Reinforcement learning (RL) ES: Aprendizaje por refuerzo · CA: Aprenentatge per reforç · EN: Reinforcement learning (RL) Learning decisions by interacting with an environment and receiving reward signals for outcomes. **An example makes it clearer** An agent practises in a game and earns points for reaching the goal, not for claiming it did. Follow the connections: AI agent, AI alignment Read more and verify: [IBM · Reinforcement learning](https://www.ibm.com/think/topics/reinforcement-learning) ## Reranking ES: Reordenación de resultados · CA: Reordenació de resultats · EN: Reranking A second ordering of retrieved results to prioritise those that best answer the query. **An example makes it clearer** Among twenty candidate lessons, another model puts the three most useful for your question first. Follow the connections: Information retrieval, Semantic search Read more and verify: [Cohere · Rerank overview](https://docs.cohere.com/docs/rerank-overview) ## Retrieval-augmented generation (RAG) ES: Generación aumentada por recuperación (RAG) · CA: Generació augmentada per recuperació (RAG) · EN: Retrieval-augmented generation (RAG) A technique that retrieves external information and supplies it to a model to prepare a contextualised answer. **An example makes it clearer** The assistant first searches the current Aula programme and then answers; the reply still needs checking. Follow the connections: Information retrieval, Grounding Read more and verify: [IBM · Retrieval augmented generation](https://www.ibm.com/think/topics/retrieval-augmented-generation) ## Robustness ES: Robustez · CA: Robustesa · EN: Robustness Ability to maintain adequate performance under variation, noise or difficult conditions within the evaluated scope. **An example makes it clearer** A transcriber should also be tested with noise, different accents and less-than-perfect microphones. Follow the connections: AI evaluation, Automatic speech recognition (ASR) Read more and verify: [NIST · AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) ## Self-supervised learning ES: Aprendizaje autosupervisado · CA: Aprenentatge autosupervisat · EN: Self-supervised learning Learning that derives training signals from the data itself, without manually labelling every answer. **An example makes it clearer** You hide part of a text and train the model to predict it from the remainder. Follow the connections: Large language model (LLM), Training Read more and verify: [IBM · Self-supervised learning](https://www.ibm.com/think/topics/self-supervised-learning) ## Semantic search ES: Búsqueda semántica · CA: Cerca semàntica · EN: Semantic search Search that considers relationships in meaning rather than only exact matches to written words. **An example makes it clearer** Searching “make a photo wider” may return an explanation of outpainting. Follow the connections: Embedding, Reranking Read more and verify: [IBM · Vector databases](https://www.ibm.com/think/topics/vector-database) ## Streaming response ES: Respuesta en flujo · CA: Resposta en flux · EN: Streaming response Progressively delivering a response as it is generated rather than waiting for the entire output. **An example makes it clearer** You see text appear gradually. An interrupted connection can leave an answer incomplete. Follow the connections: Latency, Application programming interface (API) Read more and verify: [Anthropic · Streaming messages](https://platform.claude.com/docs/en/build-with-claude/streaming) ## Supervised learning ES: Aprendizaje supervisado · CA: Aprenentatge supervisat · EN: Supervised learning Training on examples that include the reference answer the model should learn to predict. **An example makes it clearer** Each practice message is labelled “enquiry” or “issue”. Follow the connections: Label, Unsupervised learning Read more and verify: [IBM · Supervised learning](https://www.ibm.com/think/topics/supervised-learning) ## Synthetic data ES: Datos sintéticos · CA: Dades sintètiques · EN: Synthetic data Artificially created data that simulates examples. It is not automatically accurate, representative or anonymous. **An example makes it clearer** You invent orders for a fictional shop to test a screen without using real customers. Follow the connections: Data privacy, Data quality Read more and verify: [IBM · Synthetic data](https://www.ibm.com/think/topics/synthetic-data) ## System prompt ES: Instrucción de sistema · CA: Instrucció de sistema · EN: System prompt An application-level instruction setting the assistant's general behaviour. It does not replace permission controls. **An example makes it clearer** The Aula assistant receives tone guidance; data access is also restricted by the system. Follow the connections: Prompt injection, Guardrails Read more and verify: [Anthropic · Prompt engineering overview](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview) ## Temperature ES: Temperatura · CA: Temperatura · EN: Temperature A setting modifying the distribution used to choose outputs. Higher temperature usually increases variation, not truthfulness. **An example makes it clearer** To compare styles, change this setting if available; no setting eliminates errors. Follow the connections: Top-p sampling, AI hallucination Read more and verify: [Hugging Face · Generation configuration](https://huggingface.co/docs/transformers/en/main_classes/text_generation) ## Test data ES: Datos de prueba · CA: Dades de prova · EN: Test data Examples separated from development to estimate final performance. Repeatedly consulting them can compromise that estimate. **An example makes it clearer** It is the held-back exam, not the answer sheet used to prepare every adjustment. Follow the connections: Validation data, AI evaluation Read more and verify: [scikit-learn · Cross-validation and held-out data](https://scikit-learn.org/stable/modules/cross_validation.html) ## Text-to-image generation ES: Generación de texto a imagen · CA: Generació de text a imatge · EN: Text-to-image generation Creating images guided by a written description, with results that need checking against the request. **An example makes it clearer** You request a bright library for a cover; check whether the lettering on its sign is actually readable. Follow the connections: Diffusion model, Prompt Read more and verify: [Hugging Face · Text-to-image](https://huggingface.co/docs/diffusers/en/using-diffusers/conditional_image_generation) ## Text-to-speech (TTS) ES: Síntesis de voz · CA: Síntesi de veu · EN: Text-to-speech (TTS) Generating artificial speech from text. The resulting voice depends on the system and its settings. **An example makes it clearer** You turn an explanation into audio for listening. Imitating a real voice raises separate authorisation questions. Follow the connections: Automatic speech recognition (ASR), Deepfake Read more and verify: [Hugging Face · Text-to-speech](https://huggingface.co/tasks/text-to-speech) ## Throughput ES: Capacidad de procesamiento · CA: Capacitat de processament · EN: Throughput The amount of work processed per unit of time, such as requests or tokens per second. **An example makes it clearer** A service can handle many requests per minute even if each takes several seconds. Follow the connections: Latency, Rate limit Read more and verify: [Hugging Face · Inference optimisation](https://huggingface.co/docs/transformers/en/optimization_overview) ## Token ES: Token · CA: Token · EN: Token A unit processed by a model. In text it can be a word, part of a word or punctuation. **An example makes it clearer** Do not estimate cost from word count alone: languages and tokenizers can split the same text differently. Follow the connections: Tokenization, Context window Read more and verify: [Google Cloud · Generative AI glossary](https://docs.cloud.google.com/docs/ai-ml/generative-ai/glossary) ## Tokenization ES: Tokenización · CA: Tokenització · EN: Tokenization The process of splitting and converting an input into units the model can process. **An example makes it clearer** A long word may be split into several pieces; the split depends on the tokenizer used. Follow the connections: Token, Large language model (LLM) Read more and verify: [Hugging Face · Tokenizers](https://huggingface.co/learn/llm-course/en/chapter2/4) ## Tool calling ES: Llamada a herramientas · CA: Crida a eines · EN: Tool calling A mechanism for a model to request an external function; the application controls execution and returns results. **An example makes it clearer** The model requests a calendar lookup; the application checks permissions and performs the authorised query. Follow the connections: Application programming interface (API), AI agent Read more and verify: [Anthropic · Tool use with Claude](https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview) ## Top-p sampling ES: Muestreo top-p · CA: Mostreig top-p · EN: Top-p sampling Sampling from the smallest token set whose combined probability reaches a chosen threshold. **An example makes it clearer** A lower threshold usually allows fewer alternatives; it is not a “percentage of correct answers”. Follow the connections: Temperature, Token Read more and verify: [Hugging Face · Generation configuration](https://huggingface.co/docs/transformers/en/main_classes/text_generation) ## Training ES: Entrenamiento · CA: Entrenament · EN: Training Process that adjusts a model's parameters using data and a learning objective. **An example makes it clearer** Showing many labelled examples helps fit the detector; using it afterwards is a different stage. Follow the connections: Model parameter, Inference Read more and verify: [IBM · Machine learning](https://www.ibm.com/think/topics/machine-learning) ## Training batch ES: Lote de entrenamiento · CA: Lot d'entrenament · EN: Training batch A group of examples processed together in a training step, depending on the configuration. **An example makes it clearer** With batches of 16, training processes small groups rather than the whole dataset at once. Follow the connections: Epoch, Training Read more and verify: [IBM · Hyperparameter tuning](https://www.ibm.com/think/topics/hyperparameter-tuning) ## Training data ES: Datos de entrenamiento · CA: Dades d'entrenament · EN: Training data Examples used to fit the model. Their variety, quality and origin influence what it learns. **An example makes it clearer** If you train a reader only on print, handwriting may come as quite a surprise. Follow the connections: Test data, Data quality Read more and verify: [scikit-learn · Cross-validation and held-out data](https://scikit-learn.org/stable/modules/cross_validation.html) ## Transfer learning ES: Aprendizaje por transferencia · CA: Aprenentatge per transferència · EN: Transfer learning Reusing what was learned on one task to help with another, rather than starting entirely from scratch. **An example makes it clearer** You start with a model that recognises shapes and adapt it to distinguish types of plants. Follow the connections: Fine-tuning, Foundation model Read more and verify: [Hugging Face · Fine-tuning](https://huggingface.co/docs/transformers/en/training) ## Transformer ES: Transformer · CA: Transformer · EN: Transformer A neural network architecture based on attention mechanisms for relating elements in a sequence. **An example makes it clearer** When processing a sentence, it can relate distant words rather than only examining the previous one. Follow the connections: Attention, Large language model (LLM) Read more and verify: [Vaswani et al. · Attention Is All You Need](https://arxiv.org/abs/1706.03762) ## Underfitting ES: Subajuste · CA: Subajustament · EN: Underfitting The model fails to learn relevant patterns sufficiently and performs poorly even on training examples. **An example makes it clearer** An overly simple model tries to explain a complicated curve with one straight line. Follow the connections: Overfitting, AI model Read more and verify: [IBM · Overfitting](https://www.ibm.com/think/topics/overfitting) ## Unsupervised learning ES: Aprendizaje no supervisado · CA: Aprenentatge no supervisat · EN: Unsupervised learning Finding structure or groups in data without labelled target answers for each example. **An example makes it clearer** You group similar comments to discover topics; nobody has assigned each comment to a group beforehand. Follow the connections: Supervised learning, Embedding Read more and verify: [IBM · Unsupervised learning](https://www.ibm.com/think/topics/unsupervised-learning) ## Validation data ES: Datos de validación · CA: Dades de validació · EN: Validation data Examples held aside to compare settings during development, without directly using them to fit weights. **An example makes it clearer** You compare two detector versions on validation data and save the final test for the end. Follow the connections: Test data, Hyperparameter Read more and verify: [scikit-learn · Cross-validation and held-out data](https://scikit-learn.org/stable/modules/cross_validation.html) ## Vector database ES: Base de datos vectorial · CA: Base de dades vectorial · EN: Vector database A system that stores and indexes vectors to efficiently find similar items. **An example makes it clearer** You store lesson representations to retrieve those closest to a person's question. Follow the connections: Embedding, Semantic search Read more and verify: [IBM · Vector databases](https://www.ibm.com/think/topics/vector-database) ## Watermark ES: Marca de agua · CA: Marca d'aigua · EN: Watermark A visible or imperceptible signal embedded in content to provide identification or origin information. **An example makes it clearer** An image can contain an invisible mark; not seeing one does not prove it is a real photograph. Follow the connections: Content provenance, Content Credentials Read more and verify: [Google DeepMind · SynthID](https://deepmind.google/models/synthid/) ## Zero-shot prompting ES: Petición sin ejemplos · CA: Petició sense exemples · EN: Zero-shot prompting A request that describes the task without including worked examples in that request. **An example makes it clearer** “Classify this message as an enquiry or issue” requests the task without showing sample answers. Follow the connections: Few-shot prompting, Prompt Read more and verify: [Anthropic · Prompting best practices](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices)