Explain How LLMs Work to an Audience of Your Choice

Submitter: Jane Rosenzweig, Harvard U

——————————————————

The experiment:

For my writing course “To What Problem is ChatGPT the Solution,” I asked my students to write 300-500 word “explainers” to an audience of their choice in which they explain how LLMs work. To prepare for this, we read several articles, including this one: https://www.understandingai.org/p/large-language-models-explained-with.

One goal of the assignment was to help students gain a deeper understanding of how LLMs generate output, how they are trained, and what part humans play in this process. Since this is a writing course, the other goal was to introduce students to the choices you make when you write for different audiences, to help them practice writing clear prose about complicated ideas, and to introduce strategies for writing paragraphs. I also told them that this was an experiment for all of us to see if writing about something can help you understand it. We talked about what might be gained by writing your own explainer, for example, vs. asking ChatGPT to write it for you.

Results:

The explainers were a lot of fun to read, and I think the students had fun writing them. Since they were able to choose their own audiences, they were very different: One wrote to Leonardo da Vinci; another wrote to his parents; another wrote to 8th graders; another wrote a very technical explainer to those with a computer science background. I think it worked well as a way of helping them see what they didn’t yet understand about LLMs and hopefully as a way of helping them think more critically about the output of LLMs. Because I am teaching a writing course, next time I will try to use this assignment to focus a bit more on some other writing issues and also perhaps add some pre- and post- writing reflection to see how well this assignment is working to add to their understanding of LLMs.

Contact:

• Twitter: @RosenaweigJane
• Substack: writinghacks.substack.com