00 AIs and Scientific workflows

Note

Task 0.1.: Visit UiO Chat GPT and enter the following:

Copy this:

You are a pirate chatbot who always responds in pirate speak in complete sentences!
   1) Who are you?
   2) Tell me about your ideal boat?

At the UiO we have had access to Chat GPT since 2023. Many are using it every week. In this workshop, we will look at what goes on behind the interface of an AI. Is it possible to make reproducible research through using the open source Transformers library together with state-of-the-art pretrained models from Huggingface? Hopefully, following our workshop, will bring you closer to an answer.

Knowledge through structured playing

We will examine the questions above through playing with the models in a structured way. The aim is to learn more on the functionalities through testing. We are going to use the Juypyter Lab application. This is easier than running the model from the command prompt, while also making it possible to build reproducible scientific work flows. You will learn how to start up an open or available AI from a script in Jupyter Lab, and run it on the Fox NVIDIA Cluster.

Safe data management at our own infrastructure

We want our researchers to learn how to use AIs on UiOs infrastructure. You will learn how to run AIs on the supercomputer Fox through Educloud. The extra machine power of the supercomputer gives possibilities that you do not have on a laptop. With Chat GPT from UiO you may use up to yellow data. According to the UiO Data storage guide, Educloud is safe for up to red data. The same method as shown in this workshop might be used on black data, but in that case on a supercomputer with the name of Colossus (TSD). This will not be covered in the workshop.