Skip to main content
Zubnet AI › Learn › Wiki

AI Wiki — Page 7 of 10

AI concepts explained by builders, not textbooks. No jargon walls. No academic gatekeeping. Just clear, practical definitions of the terms you'll actually encounter.

454 terms 8 categories Updated 2026-11-22
💡
Term of the Day
Deepgram
Speech AI company building fast, accurate speech recognition and text-to-speech APIs. Their Nova models compete with and often beat OpenAI's Whisper…
→
No terms on this page match your filters.

301–350 of 454 terms · Page 7 of 10

Filter this page

ParametersAlso known as: Weights, Model Parameters
Fundamentals
The internal values a neural network learns during training — essentially the "knowledge" of the model encoded as…
pass@kAlso known as: pass@1, pass@10, Pass at k
Training
A code generation metric that measures the chance that at least one of k attempts by the model passes all the tests.
PEFTAlso known as: Parameter-Efficient Fine-Tuning
Training
A family of fine-tuning methods that train only a small number of extra or selected parameters instead of the whole…
PerplexityAlso known as: AI-powered search engine, Sonar API
Companies
AI search engine that combines real-time web search with language model reasoning to give direct, sourced answers…
Fundamentals
A measurement of how well a language model predicts text. Technically, it's the exponential of the average…
PhiAlso known as: Microsoft Phi
Models
Microsoft's family of small language models, ranging from roughly 1 billion to 14 billion parameters, built to deliver…
PikaAlso known as: Pika Labs
Companies
An AI video generation company whose web app turns text prompts, still images, and uploaded footage into short video…
PixVerseAlso known as: PixVerse video generation
Companies
Chinese video generation company building accessible AI video tools. Known for fast generation speeds and a free tier…
PoolingMax Pooling, Average Pooling
Fundamentals
An operation that reduces the spatial dimensions of data by summarizing a region into a single value. Max pooling…
Pose EstimationBody Pose, Skeleton Detection, Keypoint Detection
Using AI
Detecting the position and orientation of a human body (or animal, hand, face) in an image or video by locating key…
Positional EncodingPositional Embedding, ALiBi
Fundamentals
A mechanism that tells a Transformer model the order of tokens in a sequence. Unlike RNNs which process tokens…
Post-TrainingAlso known as: Post-training
Training
The set of training stages applied to a model after pretraining, turning a raw next-token predictor into a usable…
PPOAlso known as: Proximal Policy Optimization
Training
PPO (Proximal Policy Optimization) is a reinforcement learning algorithm that updates a policy in small, controlled…
Training
The initial, massive training phase where a model learns language (or other modalities) from a huge corpus. This is…
Precision & RecallF1 Score, Confusion Matrix
Fundamentals
Two complementary metrics for evaluating classifiers. Precision answers "of the items the model flagged as positive,…
Process Reward ModelAlso known as: PRM, Step-Level Verifier
Training
A model that scores each intermediate step of a reasoning chain, rather than only the final answer, so that good…
Fundamentals
The text you give to an AI model to get a response. A prompt can be a question, an instruction, a creative brief, or a…
Prompt CachingContext Caching, Prefix Caching
Using AI
A technique that saves and reuses the processed version of a prompt prefix across multiple API calls, avoiding…
The practice of crafting inputs to get better outputs from AI models. This ranges from simple techniques (being…
Prompt InjectionIndirect Prompt Injection
Safety
An attack where malicious instructions are embedded in content that an AI model processes, causing the model to follow…
Prompt TemplateTemplate, Prompt Pattern
Using AI
A reusable prompt structure with variable placeholders that gets filled in with specific data at runtime. Instead of…
PruningModel Pruning, Weight Pruning
Training
Removing unnecessary parameters (weights, neurons, or entire layers) from a trained model to make it smaller and…
PyTorchAlso known as: Torch
Tools
The open-source deep learning framework that most AI research and most open models are built with.
QLoRAAlso known as: Quantized LoRA
Training
A fine-tuning method that trains LoRA adapters on top of a model frozen in 4-bit precision, so large models fit on a…
QuantizationAlso known as: GGUF, GPTQ, AWQ
Infrastructure
Reducing a model's precision to make it smaller and faster. A model trained in 32-bit floating point can be quantized…
Question AnsweringQA, Reading Comprehension
Using AI
A system that answers questions posed in natural language. Extractive QA finds the answer span within a given document…
QwenAlso known as: Tongyi Qianwen
Models
Qwen is a family of open-weight large language models developed by Alibaba Cloud. The lineup spans several generations…
RAGAlso known as: Retrieval-Augmented Generation
Tools
A technique that gives AI models access to external knowledge by retrieving relevant documents before generating a…
Infrastructure
Restrictions on how many API requests you can make per minute/hour/day. Providers impose rate limits to prevent server…
ReActAlso known as: Reason + Act, Reasoning and Acting
Using AI
An agent pattern where a model alternates between reasoning about what to do, taking an action with a tool and reading…
ReasoningAlso known as: AI Reasoning, Chain-of-Thought Reasoning
Using AI
The ability of AI models to think step-by-step, decompose complex problems, and arrive at logically sound conclusions.…
RecraftAlso known as: Recraft V3, vector graphics generation
Companies
AI design tool focused on professional-grade image and vector graphic generation. One of the first to produce truly…
Safety
The practice of deliberately trying to make an AI model fail, misbehave, or produce harmful outputs. Red teams probe…
RegressionLinear Regression, Prediction
Fundamentals
A machine learning task that predicts a continuous numerical value rather than a category. "What will the temperature…
RegularizationAlso known as: L1 / L2 Regularization
Training
Any technique that discourages a model from memorizing its training data so it generalizes better to new data.
Reinforcement LearningAlso known as: RL
Training
A training paradigm where an AI agent learns by interacting with an environment, taking actions, and receiving rewards…
RekaAlso known as: Reka Core, Reka Flash
Companies
AI research company founded by former DeepMind, Google Brain, and FAIR researchers. Building natively multimodal…
Companies
A cloud platform for running machine learning models through a simple API, with no infrastructure to manage. It hosts…
RerankingAlso known as: Re-ranking, Cross-Encoder Reranking
Tools
A second scoring pass in a retrieval pipeline that re-orders an initial set of candidate documents with a more…
Resemble AIAlso known as: Voice cloning, speech synthesis, watermarking
Companies
Canadian voice AI company specializing in high-fidelity voice cloning and real-time speech synthesis. One of the first…
Residual ConnectionSkip Connection, Shortcut Connection
Fundamentals
A connection that bypasses one or more layers by adding the input directly to the output: output = layer(x) + x.…
ResNetAlso known as: Residual Network
Models
The 2015 image model that used skip connections to train networks over 100 layers deep, an idea now inside nearly…
Responsible Scaling PolicyAlso known as: RSP, Frontier Safety Framework, Preparedness Framework
Safety
A public commitment by an AI developer to test its models for dangerous capabilities and to have specific safeguards…
RetrievalInformation Retrieval, IR
Fundamentals
The process of finding relevant documents, passages, or data from a large collection in response to a query. In AI,…
Reward HackingAlso known as: Specification Gaming, Reward Gaming
Safety
When an AI finds a way to score well on its reward signal without actually doing what its designers intended.
Reward ModelRM, Preference Model
Training
A model trained to predict human preferences between AI responses. Given a prompt and two candidate responses, the…
RLAIFRL from AI Feedback
Training
A variant of RLHF where the preference labels come from an AI model instead of human annotators. A strong AI model…
RLHFAlso known as: Reinforcement Learning from Human Feedback
Training
A training technique where human evaluators rank model outputs by quality, and this feedback is used to train a reward…
RLVRAlso known as: Reinforcement Learning with Verifiable Rewards, RL with Verifiable Rewards
Training
A reinforcement learning method in which the reward signal comes from automatic, programmatic checkers — such as…
RNNRecurrent Neural Network, LSTM, GRU
Models
A neural network that processes sequences by maintaining a hidden state that gets updated at each step — it…
ESC