100 Words You Need to Know to Understand AI

No.	Term	Definition
1.	AI alignment	Making AI goals match human values.
2.	AI ethics	Moral principles guiding AI design and use.
3.	Algorithm	Step-by-step procedure for solving a problem.
4.	API	Interface letting software systems communicate.
5.	Artificial general intelligence	Human-level AI across many tasks.
6.	Artificial intelligence	Machines performing tasks requiring human intelligence.
7.	Attention mechanism	Model focuses on relevant parts of input.
8.	Autoencoder	Learns compressed data representations by reconstruction.
9.	Automation	Using machines to perform tasks automatically.
10.	Backpropagation	Method to compute gradients for neural networks.
11.	Bagging	Ensemble technique averaging models trained on bootstraps.
12.	Baseline	Simple reference model for comparison.
13.	Bayesian inference	Updating beliefs using probability and evidence.
14.	Benchmark	Standard test used to compare models.
15.	Bias	Systematic error causing unfair or skewed results.
16.	Big data	Very large datasets requiring scalable processing.
17.	Binary classification	Predicting one of two possible classes.
18.	Boosting	Ensemble method combining many weak learners.
19.	Chatbot	System that converses with users in text.
20.	Classification	Assigning inputs to predefined categories.
21.	Clustering	Grouping similar items without labeled examples.
22.	Computer vision	AI that interprets images and video.
23.	Confusion matrix	Table summarizing classification prediction outcomes.
24.	Context window	Amount of text a model can consider.
25.	Convolutional neural network	Neural network specialized for grid-like data.
26.	Corpus	Collection of text used for training.
27.	Cross-validation	Testing model across multiple train-test splits.
28.	Data augmentation	Creating varied training examples from existing data.
29.	Data drift	Input data changes over time, hurting performance.
30.	Data labeling	Adding human-provided tags to training data.
31.	Dataset	Organized collection of data for learning.
32.	Decision tree	Tree-structured model splitting features to predict.
33.	Deep learning	Machine learning using many-layer neural networks.
34.	Diffusion model	Generates data by denoising from noise.
35.	Dimensionality reduction	Reducing features while preserving important structure.
36.	Distributed training	Training across multiple machines or GPUs.
37.	Dropout	Regularization by randomly disabling neurons during training.
38.	Embedding	Numeric vector representing meaning or similarity.
39.	Ensemble	Combining multiple models to improve predictions.
40.	Evaluation metric	Number measuring model performance on a task.
41.	Explainable AI	Methods making model decisions understandable.
42.	Feature	Input variable used by a model.
43.	Feature engineering	Creating useful features from raw data.
44.	Few-shot learning	Learning from a small number of examples.
45.	Fine-tuning	Adapting a pretrained model to new data.
46.	Foundation model	Large pretrained model adaptable to many tasks.
47.	Generative adversarial network	Generator and discriminator trained to create realistic data.
48.	Generative AI	AI that creates text, images, or audio.
49.	GPU	Processor optimized for parallel numerical computation.
50.	Gradient	Direction and rate of change of loss.
51.	Gradient descent	Optimization method minimizing loss via gradients.
52.	Hallucination	Model outputs plausible but incorrect information.
53.	Hyperparameter	Setting chosen before training, like learning rate.
54.	Image segmentation	Labeling each pixel with a class.
55.	Inference	Using a trained model to make predictions.
56.	Instruction tuning	Training to follow human instructions better.
57.	K-means	Clustering algorithm minimizing within-cluster variance.
58.	Knowledge graph	Linked entities and relations stored as a graph.
59.	Large language model	Neural network trained to predict text tokens.
60.	Latency	Time delay to produce a model response.
61.	Learning rate	Step size used during optimization updates.
62.	Linear regression	Predicting a numeric value with a linear model.
63.	Loss function	Measure of model error to minimize.
64.	Machine learning	Algorithms learning patterns from data to predict.
65.	Model	Mathematical system mapping inputs to outputs.
66.	Model collapse	Generative model degrades from self-generated training data.
67.	Multimodal	Using multiple data types, like text and images.
68.	Natural language processing	AI techniques for understanding and generating language.
69.	Neural network	Layered function approximator inspired by brains.
70.	Normalization	Scaling features to comparable ranges.
71.	Object detection	Finding and labeling objects in images.
72.	Overfitting	Model fits training data but fails on new.
73.	Parameter	Learned weight value inside a model.
74.	Perplexity	Language-model uncertainty measure; lower is better.
75.	Precision	Fraction of predicted positives that are correct.
76.	Pretraining	Initial training on broad data before specialization.
77.	Prompt	Input text guiding a generative model.
78.	Prompt engineering	Crafting prompts to get better model outputs.
79.	RAG	Retrieval plus generation grounded in external sources.
80.	Random forest	Ensemble of decision trees using randomness.
81.	Recall	Fraction of actual positives correctly identified.
82.	Recommender system	Model suggesting items a user may like.
83.	Reinforcement learning	Learning actions via rewards from an environment.
84.	RLHF	Fine-tuning with human feedback to shape behavior.
85.	Robustness	Staying reliable under noise or adversarial inputs.
86.	Safety	Preventing harmful or unintended AI outcomes.
87.	Sampling	Choosing outputs from a probability distribution.
88.	Self-supervised learning	Learning from data using automatically generated labels.
89.	Sentiment analysis	Detecting opinions or emotions in text.
90.	Supervised learning	Learning from labeled input-output examples.
91.	Synthetic data	Artificially generated data resembling real data.
92.	Temperature	Sampling setting controlling randomness of outputs.
93.	Test set	Held-out data used only for final evaluation.
94.	Token	Basic text unit processed by language models.
95.	Tokenization	Splitting text into tokens for modeling.
96.	Training	Optimizing model parameters using data.
97.	Training set	Data used to fit model parameters.
98.	Transfer learning	Reusing learned knowledge for a new task.
99.	Transformer	Neural architecture using attention for sequence modeling.
100.	Turing test	Test of machine behavior indistinguishable from human.