TOPIC · ENGLISH EDITION
Concepts
Core technical concepts used across modern AI systems.
AI Agent
An AI system that selects and executes actions over multiple steps to pursue a goal.
AI Agent Oversight
Human control over agent goals, tool permissions, execution and outcomes, including meaningful reviews and recovery from failures.
AI Alignment
Research and design for relating AI behavior to human intentions, values and constraints, including reward proxies, feedback and operational evaluation.
AI Memory
Stored conversation state, preferences and task records, distinguished from context windows and model training, with examples of freshness and access boundaries.
Attention
A mechanism that lets a neural network weigh the relevance of different positions when computing a representation.
Context Window
The amount of tokenized input and generated text a model can consider within one inference context.
Diffusion Model
A generative model that learns to reverse a gradual noise process to create data such as images.
Embedding
A numerical vector representation used to encode semantic or structural properties of data.
Fine-tuning
Additional training that adapts a pre-trained model to a narrower task, domain, or behavior.
GPU
A highly parallel processor widely used to train and run modern machine-learning models.
Inference
The process of running a trained model on new input to produce predictions, embeddings, or generated output.
KV Cache
A cache of attention keys and values used to avoid recomputing earlier context during autoregressive generation.
Knowledge Distillation
A training method in which a smaller student model learns from the outputs or representations of a larger teacher model.
Latency and Throughput
Two core serving metrics describing response delay and the amount of work a system completes over time.
LoRA
Low-Rank Adaptation, a parameter-efficient technique for adapting large neural networks.
Mixture of Experts (MoE)
A model architecture that routes each input to a subset of specialized parameter blocks called experts.
Multimodal AI
AI systems that process or generate more than one type of data, such as text, images, audio, or video.
Physical AI
AI that connects perception and decision-making to physical actions, including robotics toolchains, sim-to-real gaps and evaluation of complete tasks.
Quantization
A technique that represents model weights or activations with lower numerical precision to reduce resource use.
RLHF
Reinforcement Learning from Human Feedback, a method for aligning model behavior with human preferences.
Retrieval-Augmented Generation (RAG)
A method that retrieves external information and supplies it to a generative model as context.
Sovereign AI
Control over AI data, infrastructure, models and operational capabilities, examined separately from data location and supplier branding.
Tokenization
The process of converting text or other input into discrete units that a model can process.
Transformer
A neural network architecture that processes sequences primarily through attention mechanisms.
Vector Search
Search that retrieves items by similarity between numerical vector representations.
Voice AI
Speech recognition, synthesis and conversation systems, including full-duplex interaction, voice replication and evaluation of actual task outcomes.