Accelerator - Specialized hardware designed to facilitate faster AI computations for higher-performance model training and inference.
AI Alignment - Effort to ensure AI systems behave in ways that align with human values and intentions.
AI Assistant - AI-powered tools designed to assist users with various tasks, such as answering questions, providing recommendations, or offering guidance.
AI Auditing - Process of evaluating AI systems for compliance with ethical standards, performance metrics, and regulatory requirements.
AI Bias - Systematic errors in AI systems that can lead to unfair or discriminatory outcomes, often reflecting prejudices in training data.
AI Governance - Policies, practices, and structures for managing the development and use of AI technologies responsibly.
AI Literacy - Understanding of AI concepts, capabilities, and limitations, enabling informed interaction with and evaluation of AI systems.
AI Risk Management - Strategies and practices to identify, assess, and mitigate potential negative impacts of AI systems.
AI Safety - Field focused on developing AI systems that are reliable, controllable, and do not cause unintended harm.
AI Alignment - Effort to ensure AI systems behave in ways that align with human values and intentions.
AI Glossary - Understanding of AI concepts, capabilities, and limitations, enabling informed interaction with and evaluation of AI systems.
AI Governance - Policies, practices, and structures for managing the development and use of AI technologies responsibly.
AI Safety - Field focused on developing AI systems that are reliable, controllable, and do not cause unintended harm.
AI Literacy - Understanding of AI concepts, capabilities, and limitations, enabling informed interaction with and evaluation of AI systems.
AI Assistant - AI-powered tools designed to assist users with various tasks, such as answering questions, providing recommendations, or offering guidance.
AI Auditing - Process of evaluating AI systems for compliance with ethical standards, performance metrics, and regulatory requirements.
AI Bias - Systematic errors in AI systems that can lead to unfair or discriminatory outcomes, often reflecting prejudices in training data.
AI Governance - Policies, practices, and structures for managing the development and use of AI technologies responsibly.
AI Risk Management - Strategies and practices to identify, assess, and mitigate potential negative impacts of AI systems.
AI Safety - Field focused on developing AI systems that are reliable, controllable, and do not cause unintended harm.
Algorithmic Fairness - Approaches to ensure AI systems make decisions without discriminating against particular groups or individuals.
Artificial General Intelligence (AGI) - A theoretical AI system that can rival and exceed the cognitive abilities of human beings across a wide variety of tasks.
Artificial Intelligence (AI) - Computer systems that can create, reason, and solve problems that would have historically required a human to perform.
Artificial Super Intelligence (ASI) - Hypothetical AI that surpasses human intelligence across all domains, raising profound ethical and existential questions.
Attention (aka attention mechanism or attention model) - Mechanism in neural networks that allows AI models to focus on specific parts of input data, improving performance on tasks like language translation and image analysis.
Back Propagation - Algorithm in machine learning that trains neural networks by calculating errors and adjusting weights backward through layers.
BERT (Bidirectional Encoder Representations from Transformers) - A transformer-based AI model that processes text by looking at both the preceding and following words for each token, aka bidirectional training.
Big Data - Extremely large datasets that require specialized AI techniques and tools for processing and analysis.
Chain of Thought - AI prompting technique that encourages language models to reason step-by-step when problem-solving, to improve output quality.
Chatbot - AI-powered tools like ChatGPT that are designed to engage in natural-sounding conversations with humans.
Chinese Room - A thought experiment challenging the notion that AI systems truly understand language, questioning the nature of machine intelligence.
Compute - Processing power and resources required for running AI tasks that influence the scale and complexity of models.
Computer Vision - Field of AI that trains computers to interpret and understand visual information from the world.
Convolutional Neural Networks (CNNs) - Specialized neural networks designed for processing grid-like data, particularly effective for image analysis tasks.
Contrastive Language-Image Pre-training (CLIP) - An NLP model that learns to connect text and images, enabling versatile visual recognition tasks.
Context Window - The amount of information an AI model can consider at once when making predictions or responses.
Conversational AI - AI systems that can interact with people in a natural, conversational manner, typically through a chat interface.
Custom Model - An AI model trained on specific data to perform a particular task, often customized for the unique needs of an individual or organization.
Data Augmentation - The process of artificially increasing the quantity or diversity of training data, helping AI models generalize better to new situations.
Data Labeling - The process of adding descriptive tags to raw data, allowing AI systems to learn from labeled examples.
Data Preprocessing - The process of cleaning and formatting raw data, making it suitable for AI model training and improving learning outcomes.
Dataset - A collection of information used to train and test AI models, enabling them to learn patterns and make predictions.
Deep Learning - A machine learning technique that uses neural networks to analyze and learn from large amounts of data.
Deepfake - AI image generation technique that replaces a person in an existing image or video with someone else's likeness.
Diffusion - AI model that generates images by gradually denoising random patterns, producing high-quality and diverse outputs.
Double Descent - A phenomenon in machine learning where model performance initially improves, then worsens and improves again as model complexity increases.
Embedding - A machine learning technique that converts discrete data into number vectors, helping AI models understand relationships between items.
Emergence - A phenomenon where complex AI behaviors arise from simple rules or interactions, often in unexpected ways.
End-to-End Learning - A machine learning technique where a model learns all steps of a task directly from input to output, without manually designed intermediate stages.
Ethical AI - Guidelines and standards for developing and deploying AI systems in ways that respect human rights and values.
Explainable AI (XAI) - Approaches that make AI decision-making processes understandable to humans, enhancing transparency and trust.
Expert Systems - AI programs that emulate human expert decision-making in specific domains, using predefined rules and knowledge bases.
Few-Shot Learning - A machine learning technique where a model is trained to achieve good performance with only a small number of training examples.
Fine-tuning - The process of refining a pre-trained AI model by training it on a smaller dataset to enhance performance on specific tasks.
Federated Learning - Approach to train AI models across multiple devices or servers, preserving data privacy by keeping raw data localized.
Forward Propagation - Algorithm in machine learning where input data flows through a neural network's layers to produce an output or prediction.
Foundation Model - Large-scale AI model trained on vast amounts of data, serving as a versatile base for various downstream tasks. Foundation models typically cost hundreds of millions of dollars to train.
Frequency Penalty - An AI parameter that discourages the model from repeatedly generating the same words or phrases.
Generative AI - AI systems that can create new content, such as images, videos, or text, based on patterns learned from its training data.
Generative Pre-trained Transformer (GPT) - An artificial neural network based on the transformer architecture that can generate human-like text by predicting the next word in a sequence.
Gradient Descent - Optimization method used in machine learning to train models by repeatedly adjusting parameters to minimize errors.
GPU - Specialized processor optimized for parallel computation to accelerate AI model training and inference.
Hallucinate/Hallucination - A phenomenon in which an AI model generates content that is not included in its training, resulting in output that's nonsensical or false.
Hidden Layer - The middle layer in a neural network that learns and represents complex features from the input data.
Hyperparameter Tuning - The process of setting the optimal configuration settings for an AI model to achieve the best performance.
Image Captioning - AI image processing technique that generates natural language descriptions of the content in images.
Image Colorization - AI image processing technique that adds color to grayscale or black-and-white images.
Image Denoising - AI image processing technique that removes noise or unwanted artifacts from images to improve their quality.
Image Generation - AI image processing technique of creating new images from scratch, often based on text descriptions or other images.
Image Inpainting - AI image processing technique for reconstructing missing or damaged parts of an image.
Image Recognition - AI image processing technique that identifies and classifies objects or scenes in digital images.
Image Segmentation - AI image processing technique that divides an image into multiple segments or objects, identifying the pixels that belong to each.
Image-to-Image Translation - AI image processing technique that takes an image input (e.g. photo) and uses it to generate a new image.
Image Upscaling - AI image processing technique that increases the resolution of an image while maintaining or improving quality.
Incremental Learning - Capability of an AI system to continuously update its knowledge from new data without forgetting previously learned information.
Inference - Once an AI model has been trained, inference is the process of using the trained model to generate new predictions or outputs.
Instruction Tuning - The process of guiding a pre-trained language model to produce higher-quality outputs by using better prompts and examples.
Large Language Model (LLM) - An AI model trained on large amounts of text, capable of understanding and generating human-like language.
Latent Space - A compressed representation of data learned by a model that captures its most important features and patterns.
Loss Function - A method for quantifying the difference between a model's predicted outputs and the actual values.
Machine Learning (ML) - A subset of AI that uses algorithms and statistics to identify patterns and extract insights from data without explicit programming.
Machine Translation - The use of machine learning algorithms to automatically translate text from one language to another.
Mixture of Experts - An AI model that combines multiple specialized models (experts) to solve different parts of a problem, with a gating network selecting the appropriate expert for a given input.
Model Compression - Techniques used to reduce the size and computational needs of AI models so they can be deployed on resource-constrained devices.
Model Deployment - The process of integrating a trained AI model into a production environment, making it available for real-world use.
Model Evaluation - Assessment of an AI model's performance using test data, ensuring it meets accuracy and reliability standards.
Model Explainability - Techniques used to understand and interpret how an AI model arrives at its decisions, promoting transparency and trust.
Model Interpretability - Techniques used to measure how easily humans can understand the reasoning behind an AI model's predictions, aiding in validation and improvement.
Model Monitoring - Continuous observation of deployed AI models to ensure consistent performance and detect potential issues or drift.
Model Training - The process of teaching an AI model to make accurate predictions by exposing it to labeled examples.
Model Versioning - System for tracking different iterations of AI models, facilitating collaboration and enabling rollback if needed.
Multimodal AI - AI systems that can process and output content in mixed data formats (e.g., text, images, audio).
Named Entity Recognition (NER) - An NLP technique that identifies and classifies named entities (like persons, organizations, locations) in text.
Natural Language Processing (NLP) - A subset of AI that focuses on enabling computers to understand, interpret, and generate human language.
NeRF (Neural Radiance Fields) - An artificial neural network that creates 3D scenes from 2D images by learning how light interacts with objects in the scene from different angles.
Objective Function - A method used in optimization problems (like machine learning) to define the goal that the algorithm aims to achieve.
Overfitting - Overfitting occurs when a model corresponds too closely to its training data and performs poorly on new data and predictions.
Parameters - The internal settings of an AI model that determine how it processes input data and generates output.
Part-of-Speech (POS) Tagging - An NLP technique that labels words in a text with their grammatical categories (e.g., noun, verb, adjective).
Presence Penalty - An AI parameter that discourages the model from generating words that have already appeared in the previous text.
Pre-training - The initial training phase of a language model on a large dataset of text and code before fine-tuning on a specific task.
Prompt - The initial text input given to an AI model, which sets the context and guides the model's output.
Prompt Engineering - The process of designing and refining prompts to extract the best responses from AI models.
Prompt Injection - The act of creating malicious AI prompts to manipulate an AI model into generating unintended or harmful outputs.
Prompt Leaking - A prompt injection attack that tricks an AI model into disclosing its system prompt, which can contain confidential or sensitive information.
Reinforcement Learning - A machine learning technique where an agent learns to make decisions by receiving rewards or penalties for actions taken in an environment.
Reinforcement Learning from Human Feedback (RLHF) - An AI technique in which human feedback is used to guide the learning process in reinforcement learning algorithms.
Regularization - An AI technique used to prevent overfitting by adding constraints to a machine learning model.
Responsible AI by Design - Approach that integrates ethical considerations and societal impact into every stage of AI system development.
RoBERTa (Robustly Optimized BERT Pretraining Approach) - An optimized version of BERT that is trained on more data for longer periods, resulting in better performance on text understanding tasks.
Sentiment Analysis - An NLP technique that determines the emotional tone or opinion expressed in a piece of text.
Singularity - Hypothetical point in time where AI surpasses human intelligence, leading to unpredictable technological and societal consequences.
Stop Sequences - Specific words or phrases that signal an AI model to stop generating text.
Style Transfer - AI image processing technique that applies the artistic style of one image to the content of another.
Supervised Learning - A machine learning technique where a model is trained on labeled data, learning to predict the output from the input.
Symbolic Artificial Intelligence - AI approach based on explicit representation of knowledge and logic-based reasoning, contrasting with machine learning methods.
Temperature - An AI parameter that controls the randomness of the output; lower temperatures make outputs predictable, while higher temperatures encourage more creative and unexpected outputs.
TensorFlow - Open-source software library for implementing and deploying machine learning models across various platforms.
Token - The smallest meaningful unit of text that an AI model can process, often representing a character or a word.
Token Limit - The maximum number of tokens a language model can process in a single input or generate in a single output.
Top-k Sampling - An AI technique that selects the next word from the top k most likely options, keeping the output more predictable.
Top-p (Nucleus Sampling) - An AI technique that selects the next word from the smallest set of most likely next words whose cumulative probability exceeds a threshold p.
TPU - Custom-built processor designed by Google for efficient execution of machine learning workloads.
Transfer Learning - A machine learning technique where a pre-trained model is adapted to a new, but related task, using its existing knowledge.
Transformer - An artificial neural network architecture that can track relationships for sequential data to transform input sequences into output sequences.
Turing Test - A test for intelligence proposed by Alan Turing, where intelligence is defined as the inability for a human to determine whether or not it is talking to another human or a computer.
Underfitting - Underfitting occurs when a model is too simple to capture the underlying relationships in the training data, resulting in poor performance on new data.
Unsupervised Learning - A machine learning technique that finds hidden patterns or intrinsic structures in input data without labeled responses.
Validation Data - A separate portion of a dataset that is not shown to the model during training in order to assess the model's performance.
XLNet - A language model that enhances context understanding by predicting words in every possible order.
Zero-Shot Learning - A machine learning technique that refers to a trained model's ability to correctly make predictions on new classes or tasks it hasn't seen before during training.
.webp)
