"Generative AI," or "Generative Artificial Intelligence," (GenAI) refers to a subset of "Artificial Intelligence" (AI) that aims to create new data that's similar to the data it has been trained on. This type of AI uses techniques like Machine Learning (more specifically Deep Learning) to analyse large amounts of data and learn to mimic or generate similar data.
Importantly, this kind of "Artificial Intelligence" is distinct from what we normally think of as AI (either "Artificial General Intelligence" or "Strong AI") which would be able to independently learn any intellectual task that a human could perform. However, GenAI is not capable of learning, or understanding the concepts underlying its own responses to human prompts.
GenAI is actually built on "Large Language Models" (LLM not AI), and these LLMs are based on the "Generative Pre-trained Transformer" architecture (or GPT—as invented by Google). These models learn patterns from a massive amount of text data (ChatGPT 3.5 was trained on 175B parameters, whilst ChatGPT-4 is trained on 1 Trillion parameters.)
LLMs use these patterns (and not logic) to generate responses (they are basically very powerful autocorrect). Unlike other computer systems that are particularly good at math, LLMs are subject to "hallucinations," where they may generate seemingly meaningful responses, that are not otherwise correct. For example, whilst a logic-based computer system will answer the prompt: 6-3= with the (correct) response: 3, an LLM may respond to the same prompt: 6-3= with the pattern-based response: 6-3=3 -> . -> 1 -> 4 and return the digit Pi. This is because the model may recognise that the more common response string is Pi (perhaps because "3.14" is more commonly seen in answers to equations rather than the simple mathematical answer.)
Because GenAI operates on a probabilistic framework, it does not possess true understanding or consciousness like a human.
This means GenAI may be a very powerful tool to help you with general concepts, planning, and structuring your work, but is not reliable to produce work by itself.
IBM Technology. (2023, July 28). How Large language Models work. YouTube. https://www.youtube.com/watch?v=5sLYAQS9sWQ