Generative Artificial Intelligence and learning in higher education

There are many definitions of what artificial intelligence is, but it is about technologies/machines that somehow act like a human, to some extent think like a human, perform actions that we associate with human intelligence, can identify patterns in complex data and is self-learning to some extent. Artificial intelligence comes in many forms and we encounter it in everyday life when we take pictures with our mobile phone's camera, get series recommendations on streaming services or suggested email replies. 

In AI, a "foundational model" is a pre-trained model that serves as a starting point for further specialization. Both generative AI and language modeling use foundational models to leverage the knowledge and insights already gained through extensive pre-training, saving time and resources when training models for specific tasks. In other words, you build on basic models to create models that are better at specific tasks. 

Generative AI is a variant of artificial intelligence that is trained to create, i.e. generate, something new such as text or images based on the data that the model has been trained on. ChatGPT, Bing AI, Perplexity.ai, Dall-E, Midjourney, and similar tools are examples of generative AI. 

A large language model (LLM) such as ChatGPT is simply an algorithm that determines how likely a sequence of words is in a given sentence and generates text based on this. A language model is generative AI and a variant of the basic model. The language models are trained on very large amounts of text (data) from different sources and then put together text based on what seems likely in response to the input/prompt given. The texts that are generated should only be seen as a text composed on the basis of probability and may contain factual errors. 

A chatbot is a computer program that has been trained with AI to mimic human oral or written conversation based on probability. There are very simple chatbots that can only handle simple questions, but also very advanced ones that can handle complex conversations. 

An automated sequence of instructions. As a recipe for baking approximately, with "if this happens" -- > "then this happens" steps in different complexity.

Machine learning involves developing algorithms and models that become better at their task through training and input to be able to make their own decisions. Input into the learning process can be human.

Deep learning is machine learning through the use of neural networks - which allows for more complex learning. Deep learning can be guided by humans, to varying degrees. Common in image classification, language processing and sound recognition.

Data is simply information that is recorded somewhere and training data is the data that is used to train an AI model through machine learning. Data can be text, music, images, and so on. 
 

Neural networks or neural networks are advanced self-learning algorithms that try to mimic the function of biological neural networks (e.g. the brain). These are an important part of deep learning in machine learning and thus also in generative AI.