How to teach ML in times of LLMs (ChatGPT, …)

[helenehoffmann]

Feel free to start a rough description of your idea as a new git issue and discuss with potential collaborators.

At the end there should be an abstract addressing the following items.

Title

An Ethical Perspective: Who is truly benefiting from AI?

How to teach ML in times of LLM (ChatGPT, …)

Description

I would like to discuss ethical questions on AI. Therefore, it would be great to have many different perspectives. I propose to focus on who is truly benefiting from it, but I'm open to shift the focus wherever the interest is.

As we teach several ML courses with hands-on coding I would like to discuss how to react to the emergence of LLMs. Small coding tasks are not useful to learn but other conceptual knowledge becomes more important.

Organizational

Organizer(s)

Me

Speakers

if applicable, In case the format foresees an introduction, short talks etc.

Format

~~In a small round I see an open discussion, but for more people a fishbowl discussion or splitting into smaller focus groups might be a good option. E.g., talks followed by open discussion, hackathon like format, world café, etc. ~~

Let’s discuss together what is really important to teach nowadays and how to do that. I propose to use the carpentries material or our open course material as use case.

Timeframe

90 min Specify the minimal time you foresee for this session, e.g., 30min, 1h, ..., up to 3h

Number of participants

minimum 3 Define the minimal number of participants you need to work on the specified task. Is there a limit to the number of people who attend this session?

[helenehoffmann]

Example material can be found here: https://deeplearning540.github.io/lesson03/content.html

[SusanneWenzel]

@helenehoffmann I understand that you don't need a screen, right? We have a few available, but possibly not for each session. Flipchart will be available

[helenehoffmann]

@SusanneWenzel I will probably be fine without a screen. Maybe we will want to look at some teaching material but the laptop might be fine for that purpose.

[SusanneWenzel]

thanks, noted

[SusanneWenzel]

@helenehoffmann if possible, please make a note here on the (rough) number of participants. Also don't forget to make a note here about the outcome of the session and, if applicable, future plans that came out of this session.

[helenehoffmann]

Number of participants: 5-8

Notes:

Do learners (and software engineers) need to know coding anymore?

• general opinion: yes
  • not to trust AI blindly
  • know how to solve bugs
  • pseudo code might suffice!?

How to recreate the learning experience of trial and error?

• is this experience really necessary?

Know how to modularize the problems!

• get help from AI for each module separately: might not work as modules are not consistent, Copilot can do that though
• needs to be taught
• also: how to prompt ChatGPT

What are the tasks we want our students to be good at?

• test and validating
  • what could go wrong!
• structure the problem into small chunks
• --> coding by AI
• understanding code

How to integrate the critical thinking learning into self-learning courses?

• includes more discussions on different perspectives
• no right and wrong anymore, difficult for automatic assessment

Assessments could include:

• find boundary cases that do not work with the generated code
• ask for the reasoning behind the solution: why did you not do this or that