How AI Works
AI Calibration
AI calibration measures whether predicted confidence levels match observed outcome frequencies over comparable cases.
Read full definitionShouldEye Trust Intelligence Glossary
How AI Works Terms
Foundational explanations of how modern AI systems process inputs, learn from data, and generate outputs.
Educational information only. ShouldEye does not provide legal, banking, or individualized financial advice. Dispute rights, deadlines, and outcomes depend on your payment method, card network, issuer, processor, jurisdiction, and the facts of your case. Legal Disclaimers
33 published definitions in this topic.
How AI Works
AI Inference Pipeline
An AI inference pipeline is the sequence of preprocessing, model computation, decoding, and postprocessing used to turn an input into a model result.
Read full definitionHow AI Works
Artificial Neuron
An artificial neuron is a mathematical unit that combines inputs with learned weights, adds a bias, and applies an activation function.
Read full definitionHow AI Works
Attention Mechanism
An attention mechanism computes context-dependent weights that determine how strongly one representation uses information from other representations.
Read full definitionHow AI Works
Decoder
A decoder converts learned representations and prior outputs into a prediction or generated sequence.
Read full definitionHow AI Works
Dense Model
A dense model generally activates the same full set of model layers and parameters for each input at a given stage.
Read full definitionHow AI Works
Emergent Capability
An emergent capability is a model behavior that appears or improves sharply as scale or training changes, especially under a chosen evaluation.
Read full definitionHow AI Works
Encoder
An encoder transforms an input into contextual numerical representations that another model component can use.
Read full definitionHow AI Works
Encoder-Decoder Model
An encoder-decoder model first represents an input and then generates or predicts an output conditioned on that representation.
Read full definitionHow AI Works
Forward Pass
A forward pass applies a model’s current parameters to input data to compute activations, predictions, and optionally a loss.
Read full definitionHow AI Works
Inference Compute
Inference compute is the computational work required to run a trained model and produce predictions or generated output.
Read full definitionHow AI Works
Latent Space
A latent space is an internal representation space in which a model encodes learned features that are not directly observed in the raw input.
Read full definitionHow AI Works
Logit
A logit is an unnormalized numerical score a model assigns to a possible output before converting scores into probabilities.
Read full definitionHow AI Works
Model Abstention
Model abstention is a system behavior that withholds a prediction or answer when evidence or confidence does not meet a defined threshold.
Read full definitionHow AI Works
Model Alignment
Model alignment is the effort to make an AI system’s behavior better match intended human goals, instructions, and constraints.
Read full definitionHow AI Works
Model Architecture
A model architecture is the organized design of layers, connections, operations, and information flow used by an AI model.
Read full definitionHow AI Works
Model Bias
Model bias is a systematic pattern in AI outputs or errors that can favor, disadvantage, or misrepresent particular cases or groups.
Read full definitionHow AI Works
Model Refusal
A model refusal is an output that declines to fulfill some or all of a request, usually because of learned or system-enforced constraints.
Read full definitionHow AI Works
Multi-Head Attention
Multi-head attention runs several learned attention operations in parallel and combines their outputs.
Read full definitionHow AI Works
Reinforcement Learning
Reinforcement learning trains an agent or policy to choose actions that increase expected cumulative reward through interaction.
Read full definitionHow AI Works
Sampling
Sampling selects an output from a model’s probability distribution, often introducing controlled variation into generation.
Read full definitionHow AI Works
Scaling Law
A scaling law is an empirical relationship describing how a model metric changes as data, parameters, or compute increase under defined conditions.
Read full definitionHow AI Works
Self-Supervised Learning
Self-supervised learning creates training targets from the structure of unlabeled data rather than requiring a person to label every example.
Read full definitionHow AI Works
Softmax
Softmax converts a set of numerical scores into a probability distribution whose values are positive and sum to one.
Read full definitionHow AI Works
Sparse Model
A sparse model uses structured or learned sparsity so only part of its parameters or connections contribute to a given computation.
Read full definitionHow AI Works
Supervised Learning
Supervised learning trains a model from examples paired with target labels or outputs.
Read full definitionHow AI Works
Tokenization
Tokenization converts raw text or other input into the discrete token identifiers a model processes.
Read full definitionHow AI Works
Top-K Sampling
Top-k sampling limits token selection at each generation step to the k candidates with the highest model probabilities.
Read full definitionHow AI Works
Training Compute
Training compute is the computational work used to optimize a model’s parameters over training data.
Read full definitionHow AI Works
Training Data
Training data is the collection of examples used to adjust a model’s parameters toward a learning objective.
Read full definitionHow AI Works
Unsupervised Learning
Unsupervised learning finds structure in data without relying on a target label for every example.
Read full definitionHow AI Works
Vector Space
A vector space is a mathematical coordinate system in which AI representations are stored and transformed as arrays of numbers.
Read full definitionHow AI Works
Vocabulary
A model vocabulary is the fixed set of token units and special symbols that its tokenizer can map to identifiers.
Read full definition