WAKIB Lexicon
One Arabic word for every term in AI. Search in either language.
78 results
78 terms10 categories
78 results
الوكيل الذكي
An AI agent is a system built on a language model that can plan, use tools, and take multi-step actions toward a goal with limited human supervision, rather than just answering a single question. Examples include agents that browse the web, write and run code, or manage a workflow.
Agents
Open termمواءمة الذكاء الاصطناعي
Alignment is the effort to make an AI system's goals, behavior, and outputs actually match what its human designers and users intend, especially as the system becomes more capable. Misalignment shows up as a model that is technically competent but pursues the wrong objective or ignores instructions in harmful ways.
Safety & Ethics
Open termسلامة الذكاء الاصطناعي
AI safety is the field of research and engineering practice focused on preventing AI systems from causing unintended harm, whether through errors, misuse, or unforeseen behavior at scale. It covers technical work like testing and guardrails as well as policy questions about deployment.
Safety & Ethics
Open termمعالجة اللغة العربية
The body of work on getting machines to process Arabic, which is harder than English for structural reasons: rich morphology, absent short vowels, and a standard written form that differs from every spoken variety.
Language & NLP
Open termتقطيع النص العربي
How Arabic text is cut into tokens, and why the same meaning costs more tokens in Arabic than in English, making Arabic requests measurably more expensive and effectively shrinking the context window.
Language & NLP
Open termالذكاء الاصطناعي العام
AGI, or artificial general intelligence, refers to a hypothetical AI system that can match or exceed human ability across essentially any intellectual task, not just narrow, specialized ones. No system today meets this bar, and experts disagree sharply on when, or whether, one will.
Business & Investment
Open termآلية الانتباه
Attention is a mechanism that lets a model weigh how much each part of the input matters when producing each part of the output, instead of treating all input equally. It is the core mechanism inside the transformer architecture.
Language & NLP
Open termالانتشار العكسي
The algorithm that assigns blame for an error backwards through a network, so every weight learns how much it contributed and in which direction to move.
Training
Open termاختبار معياري
A benchmark is a standardized test or dataset used to measure and compare how well different AI models perform on a specific kind of task, such as math problems or reading comprehension. Published benchmark scores let researchers and buyers compare models on a common basis.
Evaluation
Open termالتحيز
Bias in AI is a systematic skew in a model's outputs that unfairly favors or disfavors certain groups or viewpoints, usually inherited from patterns in its training data. It can show up as stereotyped language, unequal accuracy across demographic groups, or skewed decisions in areas like hiring or lending.
Safety & Ethics
Open termسلسلة التفكير
Chain of thought is a prompting technique where a model writes out its intermediate reasoning steps before giving a final answer, instead of jumping straight to a conclusion. This step-by-step reasoning noticeably improves accuracy on math, logic, and multi-step problems.
Language & NLP
Open termتجزئة النص
Splitting documents into passages before embedding them, so retrieval returns a relevant piece rather than a whole file. Where the cuts fall largely determines whether a RAG system works.
Language & NLP
Open termالرؤية الحاسوبية
Computer vision is the field of AI focused on getting computers to interpret images and video, such as recognizing objects, faces, or text within a picture. It powers applications like photo search, medical image analysis, and self-driving cars.
Vision
Open termنافذة السياق
The context window is the maximum amount of text, measured in tokens, that a model can consider at once when generating a response. Anything beyond that limit is truncated or dropped, which is why very long documents may need to be split up.
Language & NLP
Open termالتعلم العميق
Deep learning is machine learning using neural networks with many layers, which lets a model learn complex patterns directly from raw data such as pixels or text. It is the approach behind image recognition, speech recognition, and today's language models.
Foundations
Open termالتشكيل
Adding the short-vowel marks Arabic normally omits, so a written word has one reading instead of several. It is a prerequisite for reliable Arabic speech synthesis and a persistent source of ambiguity for text models.
Language & NLP
Open termنموذج الانتشار
A diffusion model generates images, audio, or video by starting from random noise and gradually refining it, step by step, into a coherent output. It is the technique behind most modern AI image generators, including Midjourney and Stable Diffusion.
Vision
Open termتضمين
An embedding is a list of numbers that represents the meaning of a word, sentence, or image in a way a computer can compare mathematically. Items with similar meaning end up with similar embeddings, which powers search and recommendation.
Language & NLP
Open termالحقبة التدريبية
One complete pass through the training dataset. Counting epochs is how training duration is usually described, and running too many is a direct route to overfitting.
Training
Open termقانون الذكاء الاصطناعي الأوروبي
The European Union's AI regulation, which sorts systems by risk level and imposes obligations accordingly, from transparency duties up to outright prohibition for a small set of uses.
Safety & Ethics
Open termمنصة التقييم
The software that runs a model against a suite of tests reproducibly, fixed prompts, fixed scoring, fixed conditions, so that two results can legitimately be compared.
Evaluation
Open termقابلية التفسير
Explainability is the degree to which a human can understand why an AI model produced a particular output, rather than treating it as an opaque black box. It matters most in high-stakes settings like medicine or finance, where a wrong or unexplainable decision carries real consequences.
Safety & Ethics
Open termالتعلم بأمثلة قليلة
Few-shot means giving a model a small number of examples of a task directly in the prompt so it can infer the pattern and apply it to a new case. It needs no retraining, just a handful of demonstrations included in the instruction text.
Agents
Open termالضبط الدقيق
Fine-tuning is additional training that adapts an already-trained model to a specific task or dataset. It typically needs far less data and compute than training a model from scratch.
Training
Open termالنموذج التأسيسي
A foundation model is a large model trained on broad, general data that can then be adapted, through fine-tuning or prompting, to many different downstream tasks. Large language models and large image models are the most common examples.
Models
Open termالشبكة التوليدية التنافسية
A GAN, short for generative adversarial network, pits two neural networks against each other: a generator that creates fake samples and a discriminator that tries to tell them apart from real ones. Through this competition, the generator gradually learns to produce increasingly realistic output, such as synthetic faces.
Models
Open termوحدة المعالجة الرسومية
A GPU, or graphics processing unit, is a chip originally built for rendering graphics that turns out to be extremely efficient at the parallel math operations neural networks need. Most AI training and much of AI inference today runs on GPUs rather than general-purpose CPUs.
Infrastructure
Open termالانحدار التدرّجي
The optimisation method behind almost all training: measure which way the error decreases, take a small step that way, repeat.
Training
Open termحواجز الأمان
Checks placed around a model that constrain what it may receive or emit, filters, validators, allowed-action lists, enforced outside the model rather than requested inside the prompt.
Safety & Ethics
Open termالهلوسة
A hallucination is a confident but false or fabricated statement produced by an AI model, such as an invented fact, citation, or quote. It happens because the model generates plausible-sounding text rather than verifying truth.
Evaluation
Open termالبحث الهجين
Combining keyword matching with vector similarity, so a system finds both documents that mean the same thing and documents containing the exact term you typed.
Language & NLP
Open termالاستدلال
Inference is the process of running a trained model to produce an output for a new input, as opposed to training the model. It is what happens every time a user sends a prompt and gets a response.
Infrastructure
Open termتكلفة الاستدلال
Inference cost is the money spent every time a deployed model is actually run to answer a request, usually priced per token processed and generated. Unlike the one-time cost of training, inference cost recurs with every user query and scales directly with usage.
Business & Investment
Open termكسر القيود
A prompt crafted to make a model produce output its training was meant to prevent, usually by reframing the request as fiction, translation, or a hypothetical.
Safety & Ethics
Open termالتقطير المعرفي
Knowledge distillation trains a smaller student model to mimic the outputs of a larger, more capable teacher model, transferring much of its behavior at a fraction of the size and cost. This produces compact models that are cheaper and faster to run while keeping most of the original quality.
Infrastructure
Open termذاكرة المفاتيح والقيم
Stored intermediate attention values from earlier tokens, kept so each new token does not force the model to recompute the whole sequence.
Infrastructure
Open termنموذج لغوي كبير
A large language model is a neural network trained on massive text data to predict and generate language. It can answer questions, write text, and follow instructions across many tasks without task-specific training.
Models
Open termالنموذج حَكَماً
Using one language model to score another's output. It scales evaluation far beyond what human raters can cover, and inherits the judging model's own biases while doing so.
Evaluation
Open termدالة الخسارة
The formula that scores how wrong a model's output is. Training does nothing except reduce this number, which makes its definition the real statement of what the model is being asked to do.
Training
Open termالتعلم الآلي
Machine learning is a field of AI in which a system improves at a task by learning patterns from data rather than following rules written by a programmer. It underlies most modern AI systems, from spam filters to language models.
Foundations
Open termخليط الخبراء
An architecture that holds many specialised sub-networks but activates only a few per token, so total capacity grows without proportionally increasing the cost of each request.
Models
Open termبطاقة النموذج
A short published document describing what a model is, what it was trained on, how it was evaluated, and where it should not be used.
Evaluation
Open termبروتوكول سياق النموذج
MCP, short for Model Context Protocol, is an open standard that lets AI models connect to external tools, files, and data sources in a consistent way. It replaces custom, one-off integrations with a shared protocol that any compatible model or application can use.
Agents
Open termالتحليل الصرفي
Breaking a word into its root and the affixes attached to it. In Arabic a single written word can encode what English needs a whole clause for, which makes this step load-bearing rather than optional.
Language & NLP
Open termالفصحى مقابل اللهجة
The gap between Modern Standard Arabic, which dominates written text and training corpora, and the regional varieties people actually speak. Models are typically far better at the former than the latter.
Language & NLP
Open termمتعدد الوسائط
Multimodal describes a model that can understand or generate more than one type of content, such as text, images, audio, or video, within a single system. This lets it, for example, answer a question about a photo or describe the content of an audio clip.
Vision
Open termالشبكة العصبية
A neural network is a computing system loosely inspired by the brain, made of layers of connected nodes that adjust their connection strengths during training. It is the basic building block underneath nearly all modern AI models.
Foundations
Open termنموذج مفتوح الأوزان
An open-weight model is one whose trained parameters are published for anyone to download and run, unlike a closed model that is only accessible through a paid API. This lets developers self-host, inspect, and modify the model, though the training code or data may still remain private.
Models
Open termالتخصيص المفرط
Overfitting is when a model learns the training data too closely, including its noise and quirks, so it performs well on that data but poorly on new, unseen data. It is usually fixed with more data, regularization, or a simpler model.
Training
Open termالمعلمة
A parameter is one of the internal numeric values a model learns and adjusts during training, such as a connection weight in a neural network. The parameter count, often in the billions for large models, is a rough proxy for a model's size and capacity.
Foundations
Open termالتدريب المسبق
The first and by far most expensive training stage, where a model learns general structure from a very large unlabelled corpus before any task-specific adaptation.
Training
Open termالتخزين المؤقت للتوجيه
Reusing the computed representation of an unchanged prompt prefix across requests, so a long shared instruction is paid for once rather than every time.
Infrastructure
Open termهندسة التوجيه
Prompt engineering is the practice of crafting the wording, structure, and examples in an instruction to a language model to get better or more reliable output. It includes techniques like giving examples, specifying format, and breaking a task into steps.
Agents
Open termحقن التوجيه
An attack that hides instructions inside content the model will read, a web page, a document, an email, so the model follows the attacker instead of the user.
Safety & Ethics
Open termالتكميم
Quantization reduces the numeric precision of a model's weights, for example from 16-bit to 4-bit numbers, to shrink its memory footprint and speed up inference. This usually causes a small drop in accuracy in exchange for running on cheaper or smaller hardware.
Infrastructure
Open termنموذج الاستدلال
A model trained to spend additional computation working through a problem before answering, rather than responding in a single pass.
Models
Open termالاسترجاع المُعزَّز بالاستدلال
A hybrid retrieval architecture combining chain-of-thought reasoning with external knowledge lookup to handle multi-step, complex queries.
Agents
Open termالفريق الأحمر
Deliberately attacking your own system to find failures before someone else does, for AI, probing a model for harmful output, leaks, and jailbreaks under adversarial conditions.
Safety & Ethics
Open termالتعلم بالتعزيز
Reinforcement learning trains a model by letting it take actions in an environment and rewarding or penalizing the outcomes, rather than showing it labeled examples. Over many trials, the model learns a strategy that maximizes reward, as used in game-playing AI and robotics.
Training
Open termالتعلم بالتعزيز من تغذية راجعة بشرية
RLHF is a training technique that uses human ratings of a model's responses to teach it which outputs people prefer, rather than relying only on the original training data. It is a key step in making chatbots more helpful, safe, and aligned with human expectations.
Training
Open term