AI Speak: A Glossary of Terms You Need to Know

Artificial intelligence (AI) simulates human intelligence and processes using data and computer technologies. AI has existed for years, but there’s a newfound excitement thanks to generative AI. Given a short prompt, generative AI produces text, images, audio, and video in a flash. These generative AI tools are widely available—and free in some cases—which has some people wondering how AI could change how people work. The ability to automate complex tasks is very promising, but the potential for bad actors to use AI for nefarious purposes is also a major concern.

To foster a better understanding of AI and the changes it could bring, here’s a shortlist of the most relevant AI terms and definitions to know.

Algorithm

An algorithm is a step-by-step procedure or set of instructions for a computer to perform a task, solve a specific problem, or make predictions. AI algorithms are useful for data analysis, pattern recognition, and decision-making. Some popular examples include:

• Fine-Tuning: a machine learning technique used in generative AI models.

• Large Language Model Meta AI (LLaMA): a popular chat model released by Meta AI in February 2023.

• Low-Rank Adaptation (LoRA): a model that makes it easier to train large models on smaller hardware.

Artificial Intelligence (AI)

Artificial intelligence (AI) seeks to simulate human intelligence. By feeding machines data, AI detects patterns useful for problem-solving and decision-making. AI is already used across chatbots, recommendation systems, image recognition, text generation, autonomous vehicles, and cybersecurity threat detection—to name a few examples. The capabilities of AI are evolving every day and new use cases are appearing all the time.

Bias

AI bias, also known as machine learning bias, occurs when training data used to build the model contains systemic or human prejudices. These built-in biases result in poor decision-making that exhibits discriminatory behavior and often has real-world consequences. Mitigating bias is essential for producing more equitable AI models.

ChatGPT

ChatGPT is a language model developed by OpenAI using algorithms to produce text responses to user prompts in natural language. This technology garnered widespread public attention to generative AI and language models for its ease of use and accessibility.

Learn More →

Chatbot

A chatbot is an AI-powered tool designed to simulate human conversations. Chatbots offer a wide variety of uses including customer service, information retrieval, and performing simple tasks when prompted.

Cutoff Date

A cutoff date is the designated date at which information collection and training ends. AI cannot recall information past this date without tools or functions that give it present information. ChatGPT 3.5, for instance, can’t retrieve information after September 2021.

Data Mining

Data mining is a process for determining useful information from patterns detected within large data sets. This helps with clustering and classifying, as well as detecting anomalies.

Data Validation

To maintain the integrity and reliability of the AI, the data validation process checks the accuracy, completeness, and consistency of data in a dataset.

Deepfake

Deepfakes are AI-generated images, videos, or audio designed to perpetrate a hoax or spread fake news. These differ from highly-convincing generated AI images used in movies or entertainment. Experts are concerned about the effectiveness and scalability of deepfake content to influence public perception, target businesses, or dupe individuals.

Deep Learning

By leveraging layered artificial neural networks to understand complex patterns in data, deep learning can learn very complex patterns; often with more diverse inputs such as audio, images, and documents. Deep learning is useful for image and speech recognition.

Ethics in AI

To ensure AI is deployed responsibly, the systems underpinning these solutions must prioritize transparency, accountability, and fairness in their construction and use.

EU AI Act

The EU AI Act is a regulatory framework for responsible AI deployment and respects data privacy rights.

Generative Adversarial Network (GAN)

A generative adversarial network (GAN) is a machine learning model consisting of two neural networks. The first generates data, and the second discriminates and refines that data. Through this adversarial relationship, a GAN helps an AI improve predictions, distinguish between real and fake data, and better detect patterns.

Generative AI

Generative AI is an AI technology synthesizing new content—such as text, images, or audio—by learning patterns based on training data.

Learn More →

Generative Pre-Trained Transformer (GPT)

Developed by OpenAI, these popular AI algorithms power natural language processing and generative algorithms used by tools like ChatGPT. Several iterations of this technology exist, including GPT-3, GPT-3.5, and GPT-4.

Learn More →

Google Bard

Released in spring 2023, Google Bard is a generative AI chatbot powered by an advanced large language model. The tool can understand, interpret, and respond to natural language queries. It’s even capable of including images in its responses to prompts.

Hallucination

AI can sometimes present false information. These hallucinations are unintentional but can produce inaccurate results.

Large Language Model (LLM)

Large Language Models are deep learning algorithms used for generating new content, answering prompts, providing sentiment analysis, and more. Examples include Large Language Model Meta AI (LLaMA), an open-source technology released by Meta, or GPT-3.

Machine Learning

Machine learning empowers an AI to analyze and learn from data to identify patterns, make predictions, and make decisions to autonomously improve over time.

Model

An AI model is a machine learning algorithm designed to process data, learn from it, and make predictions or generate outputs without explicit programming.

Multimodal AI

AI capable of handling input and output via several mediums (such as text, images, video, and sound) is considered multimodal.

Natural Language Generation (NLG)

Natural language generation allows an AI to produce human-like text or speech, based on a given data set.

Natural Language Processing (NLP)

Natural language processing enables an AI to understand and interpret human language, whether spoken or written. NLP helps facilitate language translation, sentiment analysis, and text classification.

Natural Language Understanding (NLU)

Natural language understanding enables an AI to extract meaning, context, and intent from written text or spoken commands. The goal of this technology is to facilitate communication between humans and AI.

OpenAI

OpenAI is the American AI company behind popular technologies, including ChatGPT and Dall-E. The company’s mission is to “ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity.” The company has boomed in popularity since the public release of ChatGPT in November 2022. ​​OpenAI was used by 21.1 million users monthly in 2022.

Prompt

A prompt is a user input an AI interprets, responds to, and is used to develop and refine a large language model for generative AI.

Pathways Language Model (PaLM)

A pathways language model (PaLM) is a 540 billion parameter transformer-based large language model developed by Google AI and powering Google Bard.

Sentiment Analysis

Sentiment analysis, also known as opinion mining, is an NLP technique used to determine the emotional tone expressed in text. By processing the text, extracting features indicating tone, and classifying these according to known examples, an AI can more appropriately understand and respond to prompts.

Synthetic Data

Synthetic data is artificial data used to train AI models while avoiding scenarios where obtaining real-world data is unnecessary or challenging due to availability or sensitivity.

Training data

Training data is an information set used to train or test machine learning models so the AI can detect patterns.

Transformer

A transformer model is an AI, deep-learning architecture used for national language processing. Transformers are the backbone of many language models because they reliably process context and handle long-range dependencies and processes.

Turing Test

Proposed by British computer scientist Alan Turing in 1950, the Turing Test demonstrates a machine’s ability to exhibit intelligent behavior indistinguishable from a human. If a human cannot reliably tell the difference between a human and a computer based on responses, then a technology passes the Turing Test.

Token

In the context of generative AI, a token is the smallest unit of text used for processing, representing individual words, subwords, or characters.

Artificial intelligence has been around for a while, but the prevalence of AI is exploding with new use cases. The automation of human-like intelligence will change the way we live and work, which offers a lot of promise. The ability to rapidly produce content, facilitate easier communication between people (and machines), and complete complex tasks autonomously will revolutionize many industries. However, there are also real concerns about the consequences of AI, including exploitation by bad actors. It's up to security professionals to think ahead to stymie bad AI and foster good AI that benefits their employees, customers, and organizations.

Protect your organization from the full spectrum of email attacks with Abnormal's behavioral AI solution. Ready to learn more? Schedule a demo today!