Tap a card to flip between quick and detailed definition.
Quick
Systems that perform tasks requiring human-like intelligence.
A subset of AI that learns patterns from data.
ML using layered neural networks.
AI systems that create new content.
Large language model for token prediction.
An input variable used by a model.
The target value a model should learn to predict.
Applying a trained model to new data.
Predicting a category.
Predicting a numeric value.
Grouping similar items without labels.
Learning from labeled examples.
Learning structure from unlabeled data.
Learning from automatically generated supervision.
Learning via actions and rewards.
Model memorizes noise in training data.
Model too simple to capture signal.
Hidden target/future information leaks into features.
Table of classification outcomes.
Fraction of correct predictions.
Share of predicted positives that are correct.
Share of actual positives captured.
Harmonic balance of precision and recall.
Mean absolute error for regression.
Root mean squared error.
Rule-based model with hierarchical splits.
Ensemble of many decision trees.
Sequential ensemble focused on residual errors.
Vector representation of semantic meaning.
Angular similarity between vectors.
Gradient-based weight update process.
Mechanism for weighting relevant context.
Neural architecture based on attention blocks.
Model processing unit for text.
Maximum token span considered at once.
Decoding parameter controlling randomness.
Nucleus sampling probability cutoff.
Fluent but unsupported model output.
Constraining answers to trusted evidence.
Retrieval-Augmented Generation.
Using examples inside prompt context.
Model response constrained to a schema.
Model invokes external APIs/actions.
Plan-act-observe iteration cycle.
Time taken to return a result.