Confused by the new AI language? Here are 25 terms you should know

The AI industry is moving fast and bringing a plethora of new terminology into the tech worker's lexicon. And with that can come confusion, with business leaders under pressure to catch up, fast.

From Agentic AI to Zero-Shot Learning, we’ve defined the 25 AI terms you should know below.

A

Agentic AI: A type of AI system that can complete a specific task without human intervention. These autonomous systems have an understanding of their context and environment and can use it to make decisions or take actions.

Agents: A type of AI designed to operate independently, making decisions and actions without direct human input.

Annual Recurring Revenue (ARR): A key metric for subscription-based businesses. It refers to how much revenue a company expects to generate from customers over a one-year period.

Anthropomorphism: The act of attributing human-like traits and qualities to AI.

B

Bayesian Networks: A type of graphical model that determines the probability of an event occurring. This allows AI models to reason in uncertain conditions.

C

Chain-of-thought Prompting: An engineering technique that guides large language models (LLMs) to solve problems via a step-by-step reasoning process — similar to how a human would.

Compute Cost: Compute or computation cost refers to the total amount of time and resources required to complete a computational task, algorithm or application. For example, for memory/RAM, this would refer to the cost of storing data temporarily during processing. For storage purposes, it would be the cost of persistent data storage.

Computer Vision: A field of artificial intelligence (AI) that enables computers to interpret and understand visual information from images and videos, similar to how the human eye and brain work.

Context Window: What an AI model can "see" and process at one time. Think of it as the model's working memory or attention span.

D

Deep Learning: A machine learning technique that uses multilayered neural networks to simulate the decision-making process of the human brain.

G

Generative engine optimization (GEO): The process of optimising content discovery and citation on generative AI systems with the aim of having content featured on AI-generated search.

H

Hallucination: An AI hallucination occurs when a generative AI model produces a response that is false, nonsensical or misleading.

L

Latency: The time delay between when an AI model receives an input or request and when it generates a response.

Large Language Model (LLM): Advanced AI systems trained on large amounts of text and information to generate human-like language.

M

Multimodal AI: An AI system that can process and generate content across multiple types of media, including text, images, audio, and video.

Multilayer Perceptron (MLP): A type of artificial neural network that combines multiple layers of neurons, including an input layer, hidden layers and an output layer. This enables it to recognise and learn complex patterns in data.

N

Neural Network: A computational model that mimics the human brain. It achieves this via a series of interconnected artificial neurons that enable it to process data, learn and understand patterns and make decisions.

Natural Language Processing (NLP): A subset of computer science and AI that helps computers understand and communicate with humans.

P

Polyglot Programming: An approach that leverages multiple programming languages to write code.

Prompt Engineering: The process of refining text inputs or prompts to guide generative AI models. This is done to influence or improve the content the model generates.

Parameters:The internal variables that AI models learn from their training data. These variables are used by the AI to make predictions and decisions. A model with more parameters typically has a more advanced ability to learn from data.

T

Temperature: The temperature of an AI model refers to the predictability of its output. Having a higher temperature means the model may generate more creative outputs but is also more likely to hallucinate.

Training Data: The dataset used to develop and teach machine learning models.

V

Vibe Coding: Coined by AI researcher Andrej Karpathy in early 2025, vibe coding refers to the emerging practice of using AI to generate code from natural language prompts.

Z

Zero-Shot Learning (ZSL):An AI technique that enables a model to recognise objects or concepts it has never seen or been trained on before.

Learn more about the AI revolution.

Have an AI startup?

Wise Business could help you save money when you raise investment, hire employees or scale your solutions globally.

Already trusted by over 700,000+ customers — ranging from sole traders to SMBs and scaleups — Wise Business powers business without borders.

With a Wise Business account, you can:

• Pay employees and vendors across borders in 160+ countries and 40+ currencies
• Get paid in 23+ currencies, including GBP, EUR and USD
• Create 9+ local account details to receive payments like a local business
• Earn a competitive return on USD, EUR and GBP with Interest*
• Create, track, and manage professional invoices for free
• Simplify operations by connecting with platforms like Amazon, Etsy, and Upwork
• Manage cash flow by connecting to accounting software, including Xero and Quickbooks
• Control outgoings with employee expense debit cards
• Use the mid-market rate on all foreign transactions
• Avoid hidden-fees and subscription costs

*capital at risk. Returns not guaranteed. See wise.com/interest for more.

Get started with Wise Business