Exit fullscreen

Ever since programs such as ChatGPT have become widely available, terms such as Artificial Intelligence (AI), Generative AI (GenAI), and Large Language Model (LLM) have quickly entered the vocabulary of our daily lives. On this page, we define these three terms together, as they all relate to each other: they represent different levels of the same concept.

Exit fullscreen

Artificial Intelligence (AI)

The European Union defines Artificial Intelligence as: "AI is the ability of a machine to display human-like capabilities such as reasoning, learning, planning and creativity.AI enables technical systems to perceive their environment, deal with what they perceive, solve problems and act to achieve a specific goal. The computer receives data - already prepared or gathered through its own sensors such as a camera - processes it and responds.

AI systems are capable of adapting their behaviour to a certain degree by analysing the effects of previous actions and working autonomously." ( What is artificial intelligence and how is it used? )

Artificial intelligence as a field started as early as the 1950s, and since then many AI technologies have been created. Think of self-driving cars, systems that analyze large amounts of data to detect fraud, or systems that calculate the most optimal cultivation methods in agriculture and horticulture. Many of us have been using AI technologies in our daily life through things such as route planners, song and video suggestions on our social media apps, and spell checker in our word processors.

Generative AI (GenAI)

Generative AI is the term used for a subset of AI technologies that aim to create new things that previously did not exist, such as text, music, images, or video. In other words, these technologies have a generative function.

With this distinct goal it differentiates itself from predictive AI, which focuses on predicting future events on the basis of current data (i.e., weather predictions), and from descriptive AI, which focuses on identifying patterns and summarizing information in data (i.e., automated suggestion for key words).

Generative AI has seen a large increase in popularity in recent years due to breakthroughs in how AI models are trained and function. Well known examples of Generative AI are GPT4, by the developers of ChatGPT, the French Mistral or image generator Midjourney. But there are a lot more tools, even open source, and integrations within existing applications.

Large Language Model (LLM)

Large Language Models (LLM) are a type of Generative AI technologie that are trained on enormous amounts of language data for the equivalent of hundreds of human years (possible because of the parallel processing of data), and as a result are able to generate human-like text.

Instead of copying and pasting existing sentences, an LLM generates new texts. It does this by predicting the most probable next word in a sentence, based on texts it has been trained on. It takes context into account. In this way it can 'write' an email, generate summaries, make translations or 'answer' questions in a way that sounds human.

It is essential to realise that an LLM can only predict words. It has no knowledge. If we type the question: "Who was the first man on the moon?", we actually ask: "Given the statistical distribution of words in the large public corpus of texts, which words are most probable to follow the phrase 'Who was the first man on the moon?'". The fact that the answer returned is "Neil Armstrong", is not knowledge, but a prediction, though a lot of people mistakenly interpret this as knowledge.

It becomes even more clear if we ask "Give me a random number between 1 and 50". Surprisingly often you will get 27 as an answer. Apparently the combination "random number between 1 and 50" and "27" appears frequently in the training data. so the model predicts that "27" is the most probable answer.

Chatbots

ChatGPT, Gemini, Mistral, Claud, etc. are not language models themselves, but tools (chatbots) which allow for interaction with the underlying language model. As end user this is done by means of providing the tool with a prompt, or in other words: a question, instruction or data that you want processed.