Glossary

Activation Engineering

Learn about Activation Engineering, an inference-time technique for steering AI model behavior by modifying internal neural activations.

Agentic Economy

Explore the Agentic Economy, an emerging digital ecosystem where autonomous AI agents perform commerce and run business operations on our behalf.

Attention Sink

Learn about attention sinks in large language models and how preserving early tokens enables stable, infinite-horizon streaming generation.

Agent Harness

Learn about the agent harness, the vital software infrastructure that connects AI models to external tools, memory, and security controls.

Agent Runtime

Discover how an agent runtime environment secures, manages, and scales autonomous AI agent workflows in production enterprise settings.

Agentic workflow

Discover how agentic workflows allow AI models to autonomously plan, execute, and adapt complex multi-step processes.

Active Inference

Understand Active Inference, a physics-based framework that unifies perception, learning, and action by minimizing variational free energy.

AI Red Teaming

Discover AI Red Teaming, an essential adversarial testing process to identify and patch security vulnerabilities in AI models before production.

AgentOps

Explore AgentOps, the emerging methodology for building, monitoring, debugging, and governing autonomous AI agents in production environments.

AI Slop

Understand AI Slop, the low-quality content produced by AI models flooding digital platforms.

Autonomous AI

Learn about Autonomous AI, systems capable of operating with minimal human oversight and making decisions in complex environments.

AI alignment

Understand AI alignment: ensuring AI systems' behaviors align with human values and ethical standards to prevent unintended consequences.

Agentic systems

Learn about agentic systems: AI models capable of autonomous decision-making and collaboration to achieve complex goals.

Artificial Intelligence

Learn about Artificial Intelligence, its definition, importance, and applications in various industries.

No terms match

Try a different letter

Context Compaction

Learn about context compaction, the algorithmic pruning and summarizing of session history to fit AI agent state inside fixed token windows.

Contrastive Preference Optimization

Discover Contrastive Preference Optimization (CPO), an alignment method for optimizing model outputs using contrasted response pairs.

Corrective Retrieval-Augmented Generation

Discover how Corrective Retrieval-Augmented Generation (CRAG) evaluates and refines retrieved data to reduce AI hallucinations.

Confidential AI

Discover Confidential AI, the technology that uses hardware-secured enclaves to keep models and sensitive data encrypted while actively in use.

Constitutional AI

Learn about Constitutional AI (CAI), an alignment method that trains AI models using structured principles rather than manual human feedback.

Context Engineering

Discover context engineering, the system-level practice of structuring and optimizing runtime information to feed accurate context to AI models.

Compound AI System

Learn about Compound AI Systems and how modular architectures outperform single, monolithic AI models.

Diffusion Transformer

Learn about Diffusion Transformers (DiT), a scalable architecture combining transformer self-attention with latent diffusion models.

Direct Nash Optimization

Understand how Direct Nash Optimization (DNO) aligns AI models using game theory and general preferences rather than scalar rewards.

Disaggregated Inference

Learn how Disaggregated Inference splits LLM prefill and decode workloads across separate hardware to optimize throughput and cost.

Direct Preference Optimization

Learn about Direct Preference Optimization (DPO), a highly efficient algorithm for aligning large language models with human preferences.

Data Science

Learn about Data Science, its definition, significance, and how it is applied in various sectors.

Deep Learning

Discover what Deep Learning is, its significance, and how it is transforming various industries.

No terms match

Try a different letter

FlashAttention

Explore FlashAttention, an IO-aware exact attention algorithm that speeds up LLM training and inference by minimizing memory bottlenecks on GPUs.

Flow matching

Explore flow matching, the highly efficient generative modeling framework replacing diffusion models for faster, high-quality image generation.

Frugal AI

Learn about Frugal AI, an emerging sustainability framework aimed at reducing the computing costs and environmental impacts of artificial intelligence.

Guardrails

Discover how Guardrails provide critical runtime validation, safety, and compliance layers for generative AI applications.

GPU Orchestration

Learn about GPU Orchestration, the technology that dynamically schedules and manages compute resources to maximize efficiency and reduce AI costs.

Group Relative Policy Optimization

Discover Group Relative Policy Optimization (GRPO), an efficient RL algorithm by DeepSeek that trains reasoning models without a critic model.

Graph Retrieval-Augmented Generation

Discover Graph Retrieval-Augmented Generation (GraphRAG), a technique linking structured knowledge graphs with text generation.

Generative Engine Optimization

Learn about Generative Engine Optimization (GEO), a strategy to ensure brand representation in AI-generated content.

No terms match

Try a different letter

In-Context Learning

Learn about In-Context Learning (ICL), the mechanism enabling LLMs to learn tasks on the fly from examples inside the prompt context.

Inference-time scaling

Discover how inference-time scaling improves AI reasoning by allowing models to allocate more computational resources at query time.

Joint Embedding Predictive Architecture

Explore Joint Embedding Predictive Architecture (JEPA) and how it trains non-generative world models to understand abstract physical realities.

Key-value cache

Explore how a key-value cache (KV cache) reduces latency and saves computational resources in modern transformer models.

Knowledge distillation

Learn about Knowledge distillation, a process that simplifies and compresses large AI models into smaller, faster versions.

Kahneman-Tversky Optimization

Explore Kahneman-Tversky Optimization (KTO), a prospect-theory alternative to RLHF that simplifies LLM alignment with binary desirability feedback.

Late Interaction

Learn about Late Interaction, a hybrid retrieval paradigm that preserves token-level semantic details for faster, high-precision search.

Large Reasoning Models

Learn about Large Reasoning Models (LRMs) and how they use step-by-step deliberation, planning, and verification to solve complex problems.

Liquid Neural Network

Learn about Liquid Neural Networks (LNNs), an adaptable and highly efficient post-transformer AI architecture for processing sequential data.

LLM-as-a-Judge

Learn how LLM-as-a-Judge uses highly capable language models to autonomously evaluate and score the outputs of other generative AI applications.

Low-Rank Adaptation

Learn how Low-Rank Adaptation (LoRA) reduces AI model training costs by only updating a small percentage of model parameters.

Model Swarms

Learn about Model Swarms, an innovative framework that adapts language models through collaborative search in a high-dimensional weight space.

Model Collapse

Discover what model collapse is, why it occurs in generative AI, and its implications for synthetic data training.

Multi-head Latent Attention

Understand Multi-head Latent Attention (MLA), the memory-saving attention mechanism that powers next-generation long-context LLMs.

Multi-agent orchestration

Learn about Multi-agent orchestration, the architectural framework used to coordinate multiple AI agents in collaborative workflows.

Matryoshka Representation Learning

Discover Matryoshka Representation Learning (MRL), a technique for creating scalable, variable-length vector embeddings.

Machine Unlearning

Discover machine unlearning, the process of removing private or copyrighted training data from trained AI models without rebuilding them.

Model Quantization

Discover how Model Quantization compresses artificial intelligence networks for faster deployment and reduced memory requirements.

Mixture of Depths

Learn about Mixture of Depths, an optimization technique that dynamically coordinates token processing pathways to maximize compute efficiency.

Model Merging

Discover Model Merging, an efficient technique to combine multiple specialized AI models into one unified model without additional training cost.

Mixture of Experts

Explore Mixture of Experts (MoE), a sparse neural network architecture optimizing performance and reducing AI inference and training costs.

Model Context Protocol

Learn about the Model Context Protocol, the open-source standard enabling AI assistants to connect securely to external tools and database systems.

Mechanistic Interpretability

Explore Mechanistic Interpretability, the study of understanding AI models' internal mechanisms to ensure behavior aligns with human intentions.

Multimodal AI

Discover Multimodal AI, systems that integrate and interpret various data forms to enhance understanding and decision-making.

Machine Learning

Discover what Machine Learning is, its significance, and how it is applied across different industries.

Neuro-symbolic AI

Learn about Neuro-symbolic AI, the hybrid paradigm combining neural networks with rule-based logic for explainable, robust reasoning.

Natural Language Processing

Learn about Natural Language Processing, its definition, significance, and applications in technology.

Outcome Reward Model

Discover how Outcome Reward Models (ORMs) align large language models by evaluating and scoring complete generation paths based on the final result.

One-Bit Large Language Models

Discover One-Bit Large Language Models, an ultra-efficient architecture that drastically lowers AI memory footprints using ternary weights.

Odds Ratio Preference Optimization

Learn about Odds Ratio Preference Optimization, a unified alignment method that simplifies fine-tuning without needing a reference model.

Parameter-Efficient Fine-Tuning

Learn about Parameter-Efficient Fine-Tuning (PEFT), a cost-effective way to customize AI models without full retraining.

Physical AI

Explore Physical AI, the technology enabling autonomous systems to perceive, reason, and directly interact with the physical world.

Prompt Caching

Learn how prompt caching stores pre-computed prompt states to slash LLM costs, minimize latency, and accelerate time to first token.

Process Reward Model

Learn how Process Reward Models (PRMs) provide step-level supervision to improve AI reasoning accuracy and safety.

No terms match

Try a different letter

Recursive Self-Improvement

Learn about Recursive Self-Improvement (RSI), the process of AI autonomously rewriting and optimizing its own codebase to enhance its intelligence.

Retrieval-Interleaved Generation

Explore Retrieval-Interleaved Generation (RIG), an AI framework that fetches real-time external data dynamically during the generation process.

Reinforcement Learning from AI Feedback

Understand Reinforcement Learning from AI Feedback (RLAIF), a scalable alternative to RLHF that aligns AI using other models.

Reasoning tokens

Discover how reasoning tokens enable AI models to perform internal deliberation and scale inference-time compute for complex tasks.

Reinforcement Learning with Verifiable Rewards

Discover how Reinforcement Learning with Verifiable Rewards (RLVR) uses objective rule-based checkers to train reliable, high-reasoning AI models.

Ring Attention

Master Ring Attention, a novel parallelization method helping language models process extremely large context windows.

Representation Engineering

Discover Representation Engineering (RepE), a top-down methodology used to inspect and manipulate neural network activations to control AI behavior.

Retrieval-Augmented Fine-Tuning

Discover Retrieval-Augmented Fine-Tuning (RAFT), a technique that trains language models to extract accurate answers from retrieved documents.

Subagent

Understand subagents, specialized AI agents delegated by parent orchestrators to execute bounded tasks inside isolated context windows.

Speculative RAG

Discover Speculative RAG, a dual-model framework that accelerates and improves retrieval-augmented generation using parallel drafting.

Structured Outputs

Learn about Structured Outputs, an AI decoding method that guarantees model responses strictly follow predefined formats like JSON schemas.

Self-correction

Discover how self-correction allows AI models to autonomously identify and fix errors during generation for highly accurate reasoning.

Self-Play Fine-Tuning

Learn how Self-Play Fine-Tuning (SPIN) enables language models to improve themselves by playing against previous iterations.

Sparse autoencoders

Learn how Sparse Autoencoders (SAEs) translate dense neural network activations into sparse, human-interpretable concepts for model safety.

Semantic Routing

Explore semantic routing, a method of directing queries to specialized AI models based on semantic nuance to reduce latency and cost.

Selective State Space Models

Explore Selective State Space Models, a high-efficiency alternative to Transformers that processes long sequences with linear scaling.

Synthetic Data Generation

Explore Synthetic Data Generation, a method for generating highly realistic artificial datasets to train and evaluate AI models safely.

Speculative Decoding

Discover Speculative Decoding, a drafting technique that significantly accelerates deep learning model inference without sacrificing quality.

Sovereign AI

Learn about Sovereign AI and how countries develop independent systems to protect data privacy, culture, and national security.

Small Language Models

Discover Small Language Models: compact AI models optimized for efficient language processing on resource-constrained devices.

Test-Time Training

Understand Test-Time Training (TTT), an adaptive AI technique that updates model parameters at inference to overcome domain shifts.

No terms match

Try a different letter

Vertical AI

Understand how Vertical AI delivers deep, domain-specific intelligence for specialized industries like healthcare and finance.

Vision-Language-Action Model

Explore Vision-Language-Action (VLA) models, the multimodal neural networks powering the next generation of generalist autonomous robots.

Vibe coding

Discover vibe coding, the conversational software development practice where AI handles code generation while developers guide the system.

World Model

Explore what World Models are in AI, how they simulate physical environments, and their importance in robotics and autonomous systems.

No terms match

Try a different letter

No terms match

Try a different letter

No terms match

Try a different letter

The Language of AI