AI Terminology

Mar 2, 2024 | General

Blog

Common Terms in AI

Any new technology can be confusing, especially in a technical field where a new acronym is invented every day. While this is not a complete list, nor do we want to promote specific companies and products, the following are common industry terms and a few companies/products that are often cited because of their industry-leading position.

 

  • Activation function: A mathematical function used in neural networks to determine the output of a neuron based on its weighted inputs.
  • AI (Artificial Intelligence): The field of computer science focused on simulating intelligent behavior in machines. It encompasses a wide range of techniques, including machine learning and deep learning.
  • API (Application Programming Interface) acts as a messenger between different software programs, allowing them to communicate and exchange data with each other. It defines a set of rules and protocols that specify how one program can request information or functionality from another program.
  • Backpropagation: An algorithm used in training neural networks to adjust the weights of connections based on the error between the predicted and actual outputs.
  • Bias: A constant value added to the weighted sum of inputs in a neuron, influencing its activation.
  • Chatbot: A computer program or AI system designed to have conversations with humans, usually through text or voice interactions.
  • ChatGPT: An AI chatbot developed by OpenAI, known for its ability to hold conversations and generate human-like text.
  • Claude: An LLM from the company Anthropic, specifically designed to be more harmless and helpful than typical language models.
  • Deep learning: A powerful subset of machine learning using complex, multi-layered artificial neural networks to learn intricate patterns from massive data.
  • Embeddings: Mathematical representations of words or concepts that capture their semantic meaning. These numerical vectors allow AI models to understand relationships between different pieces of data.
  • Fine-tuning: Adapting a pre-trained model to a specific task or domain by training it on a smaller, more relevant dataset.
  • Generative AI: Subset of AI that focuses on creating new content, including text, images, music, code, and more.
  • Generative Pre-trained Transformer (GPT): A type of neural network architecture used in many LLMs, including ChatGPT.
  • Gemini: A factual language model family created by Google AI, designed to prioritize accuracy and reliability.
  • Hallucination: An incorrect or misleading output generated by an AI model due to insufficient training data, biases in the data, or other limitations. 
  • Index(ing): The process of organizing and storing data in a way that optimizes its retrieval for search or query purposes.
  • JSON (JavaScript Object Notation): A lightweight, human-readable format for data exchange, often used for transmitting data between applications.
  • Langchain: A Python library simplifying the process of building applications using large language models.
  • LLaMA: A series of open-source, foundational language models from Meta with varying sizes.
  • LlamaIndex: A project and dataset that aims to thoroughly evaluate and document open-source large language models.
  • LLM (Large Language Model): A powerful AI model trained on massive amounts of text data to communicate, generate different forms of creative text, and answer questions in an informative way.
  • Loss function: A function that measures the difference between the predicted and actual outputs, used to evaluate the performance of a model during training.
  • Machine learning: A subfield of AI where algorithms learn from data without explicit programming.
  • Mistral: An LLM developed by GoogleAI, similar to ChatGPT.
  • NLP (Natural Language Processing): Subfield of AI concerned with the interactions between computers and human language. It focuses on tasks like text understanding, language translation, and text generation.
  • OpenAI: A research company dedicated to advancing artificial intelligence safely and beneficially. Founders of popular LLMs like ChatGPT and DALL-E.
  • Parameter: A configurable value within a model that influences its behavior and is adjusted during training.
  • Prompt: The initial input or instruction given to a generative AI model, directing it to perform a task or generate a specific kind of output
  • Query: The act of asking a question or requesting information, often used in the context of searching for something on the internet or interacting with a database.
  • RAG (Retrieval-Augmented Generation): Techniques that combine information retrieval (finding relevant documents) with text generation capabilities for more informative and accurate language models.
  • Reinforcement learning: A machine learning technique where an agent learns through trial and error, receiving rewards for desired behavior.
  • Structured Data: Data organized in a predefined format (e.g., tables, databases) that is easily understandable and searchable by machines.
  • Supervised learning: A machine learning approach where the training data includes both the input and the desired output, guiding the model’s learning.
  • Temperature (Generative AI): A hyperparameter controlling the randomness of model outputs, ranging from predictable (low) to creative (high) but potentially inaccurate.
  • Transformation: The act of modifying or converting data from one format to another. This is commonly used to prepare data for use in AI models.
  • Unstructured Data: Data without a predefined structure (e.g., text, images, videos), requiring more complex processing techniques for analysis.
  • Unsupervised learning: A machine learning approach where the training data only consists of input data, and the model identifies patterns or structures without pre-defined labels.
  • XML (Extensible Markup Language): A markup language defining a set of rules for encoding documents in a format readable by both humans and machines.

Diagram of Select Terms

While this is not a complete list of the above terms, we thought it would be helpful to see some of these terms in a diagram, specifically related to RAG and Llamaindex, since that is what is used at Vizaport. Thanks to the team at Llamaindex for the diagram. You can visualize how data is fed into a RAG index, queried by a user with a prompt, and its response.

 

AI Services and Web Development Agencies

Recent Articles

Our new product line

Our new product line

Introducing Vizaport's New Suite of AI-Powered Data Visualization Tools In today’s rapidly evolving technological landscape, data visualization is not just a tool but a necessity for businesses across various sectors. At Vizaport, we are proud to unveil our redesigned...

Network Visualizer – Configuration Options

Network Visualizer – Configuration Options

New Options for Internet Service Providers In Vizaport's latest release, we've added new configuration options for our Network Visualizer, used by Internet Service Providers to allow customers to view proximity to fiber and cable lines. We'll highlight all of the...

Vizaport AI Chatbot – Now With Google Gemini

Vizaport AI Chatbot – Now With Google Gemini

From Zero to Chatbot in 3 minutes…that’s all it takes with Vizaport’s latest WordPress Plugin update. Version 1.2 now comes with an option to choose Google Gemini. You still have the option to use OpenAI, but choosing Gemini bypasses the requirement of obtaining a...