How GenAI Works

Generative AI uses a class of neural network-based algorithms that are capable of producing new and diverse content after being trained on extensive data sets, be it text, images, audio, or video.

GenAI, a type of AI, creates new content — such as text, images, music, video, or code — in response to natural language prompts. Unlike traditional search engines that retrieve existing content, GenAI produces original outputs by analysing patterns in large datasets from the internet. It mimics human-like expression but lacks true understanding of the real world or social context. While its outputs can appear fluent and convincing, they are not always accurate or reliable, e.g. ChatGPT may generate plausible sounding but incorrect or misleading information.

While there are different types of systems, common features across most GenAI tools are:

• They use machine learning, especially deep learning.
• They are trained on large datasets to learn patterns.
• They generate new content (text, images, audio, video, code, etc.) based on learned patterns.
• They rely on parameters (like ‘weights’) to guide their outputs.

Large Language Models (LLMs)

Text-based GenAI, like ChatGPT, are commonly used and powerful GenAI tools that are used in a variety of settings including research, and are based on a type of ANN called a Large Language Model (LLM), specifically a Generative Pre-trained Transformer (GPT). 

These models improve through iterations, using vast datasets and increasing computational power, with each version enhancing performance by training on more data and refining billions of parameters. The rapid progress in GenAI is driven by advances in algorithms, data, and computing power. LLMs can vary in terms of training techniques, specialisation, data sources and model, size and scope.

Typically, based on instructions (also known as prompts) provided by the user, the quality of the output produced by these models is such that it is difficult to distinguish it from human-generated content and thus its use must be approached with care to ensure ethical, accurate, and responsible application.

Discover More

For a deeper dive into the techniques behind GenAI and how LLMs generate text, see Appendix B.