Glossary of AI Terms
Zero-Shot Learning: A model’s ability to perform a task without specific training data for that task.
Artificial General Intelligence (AGI): Hypothetical AI that performs as well as or better than humans on a wide range of cognitive tasks. Also known as strong AI. Contrasts with narrow or weak AI.
AI Ethics: Principles guiding the development and use of AI to prevent harm, addressing issues like data collection and bias.
AI Safety: An interdisciplinary field concerned with the long-term impacts of AI, particularly the potential risks of advanced AI and superintelligence.
Algorithm: A set of instructions that enable a computer program to learn and analyze data, identify patterns, and perform tasks.
Alignment: Adjusting an AI model’s output to better match desired outcomes, including content moderation and human interaction.
Anthropomorphism: Attributing human-like qualities to non-human objects. In AI, this can involve believing a chatbot is more sentient or aware than it is.
Bias (in LLMs): Errors stemming from training data that can lead to skewed or discriminatory outputs, such as stereotyping.
Chatbot: A program designed to simulate human conversation through text.
ChatGPT: An AI chatbot developed by OpenAI using large language model technology.
Cognitive Computing: Another term for artificial intelligence.
Data Augmentation: Expanding training data by remixing existing data or adding new, diverse data.
Deep Learning: A subfield of machine learning that uses multiple layers of parameters (artificial neural networks) to recognize complex patterns in data (images, sound, text).
Diffusion: A machine learning method that adds random noise to data (e.g., a photo) and trains networks to reverse this process, effectively recovering the original data.
Emergent Behavior: Unintended abilities exhibited by an AI model.
End-to-End Learning (E2E): A deep learning approach where a model learns to perform a task from start to finish without sequential training.
Ethical Considerations (in AI): Awareness of the moral implications of AI, including issues of privacy, data usage, fairness, and safety.
FOOM (Fast Takeoff/Hard Takeoff): The idea that the rapid development of AGI could happen so quickly that it might be impossible to control.
Generative AI: AI that creates new content, such as text, video, code, or images, by learning patterns from large datasets.
Generative Adversarial Networks (GANs): Generative AI models using two neural networks (a generator and a discriminator) to create new data.
Google Bard: An AI chatbot by Google similar to ChatGPT but connected to the current web.
Guardrails: Policies and restrictions implemented to ensure responsible data handling and prevent AI from generating harmful content.
Hallucination (in AI): When an AI provides an incorrect or fabricated response, presented confidently as fact.
Large Language Model (LLM): An AI model trained on massive text datasets to understand and generate human-like language.
Machine Learning (ML): A branch of AI enabling computers to learn and improve predictions without explicit programming, often using training sets.
Microsoft Bing: A search engine using ChatGPT-like technology for AI-powered search results.
Multimodal AI: AI that processes multiple input types, such as text, images, video, and speech.
Natural Language Processing (NLP): A field of AI using ML and deep learning to enable computers to understand human language.
Neural Network: A computational model inspired by the human brain, composed of interconnected nodes (neurons) for pattern recognition and learning.
Overfitting: A machine learning error where a model performs too well on training data but poorly on new data.
Parameters: Numerical values that define the structure and behavior of LLMs, enabling predictions.
Prompt Chaining: AI’s ability to use information from previous interactions in subsequent responses.
Stochastic Parrot: An analogy for LLMs, highlighting their ability to convincingly mimic human language without true understanding.
Style Transfer: Applying the artistic style of one image to the content of another.
Temperature (in LLMs): A parameter controlling the randomness of a language model’s output. Higher temperature = more risk-taking.
Text-to-Image Generation: Creating images from text descriptions.
Training Data: Datasets used to train AI models (text, images, code, data).
Transformer Model: A neural network architecture that learns context by analyzing relationships within data.
Turing Test: A test of a machine’s ability to exhibit human-like behavior, passing if a human judge cannot distinguish its responses from a human’s.
Weak AI/Narrow AI: AI designed for specific tasks, unable to learn beyond its defined skill set. Most current AI is weak AI.