Generative AI for Everyone

I've been going through Andrew Ng's Coursera course titled Generative AI for Everyone to help think through how to integrate AI concepts into courses I teach. It is intended to take 2-3 weeks to complete. As with most MOOC providers, you can go through the course for free, but if you want to take the quizzes/tests and obtain a certificate of completion at the end, it costs a bit. The following notes come from going through the first week.

Generative AI systems produce high-quality text, images, audio, and video.

Supervised learning is used for labeling things. For example, when an email comes in, an AI spam filter could label that email as spam or not. Or for self-driving cars, inputs from the car's sensors can understand the vector of other cars.

A Large Language Model (LLM) can take a prompt and use it to generate output. Supervised learning outputs the next word based on the first part of the sentence as the input.

An LLM can be used to answer simple knowledge questions or as a writing partner to help refine existing text or to generate new text based on the prompt. It can also be used to create something that has not been already widely done on the internet.

AI is a general-purpose technology, meaning it can be useful for many tasks rather than being limited to only specific situations.

Interfaces may be web-based, similar to how a web search functions, or application-based, where the AI function is integrated into another application.

When considering whether an LLM could do a certain task, consider if a fresh college grad could likely complete the task.

They are limited by knowledge cut-offs where certain information has not been used to train the LLM or hallucinations where it makes up information that it doesn't know anything about. There also may be limitations as to the input and output length.