GPT This! a writing assignment in the age of GPT-3 & other Large Language Models
(update: Want more on ChatGPT, read our collaborative memoir: Hallucinate This!)
Need an assignment that is ChatGPT-proof? Look somewhere else. I’m not sure one exists, but here’s a model for a kind of writing assignment that might at least use the program in a way that stimulates — rather than replaces — student critical thinking.
For your first paper, you will be focussing on your future profession and its relationship to the communities it impacts. You have already begun answering this question in your diagnostic and in your blog post. However, we are also living in the age of programs that can write these essays for us. Or can they? In this paper assignment, you’ll reflect on the key element that an AI text generator leaves out of your essay.
Prompt: When you ask ChatGPT (or equivalent LLM text generator) to write an essay about [*the ways your future profession could better engage the communities it impacts*], what key component does the output omit?
[ * Fill in the blank with the topic you’d like them to write about
This is ultimately not really a paper ABOUT ChatGPT but one that uses the output of ChatGPT (or other LLM) to help you discover and recognize the complexities and nuances you are bringing to this topic or question.
To write this paper, follow these steps:
- Run your version of the prompt through ChatGPT or equivalent.
- Analyze what it offers and what is missing.
- Include the ChatGPT response as an attachment
On the one hand, this paper is an exercise in reflection about your chosen profession’s relationship to communities. On the other hand, this paper asks you to do some close reading, analyzing closely the textual output of the AI. Be careful. It is easy for this paper to turn into a laundry list of smaller details that are left out of ChatGPT’s output. While those details are important, put them in the service of a larger or central idea.
On using ChatGPT:
Since ChatGPT added a pay tier and can get tied up, consider another: OpenAI playground, You.com, or other.
Also, anyone can outrun a robot that has been programmed to run slow. To make this exercise as useful as possible, you should keep revising your prompt for ChatGPT until you get output that is fairly sophiticated an interesting. You can see my experiments in the readings below. Also, try to stay away from obvious points about ChatGPT, such as its inability to use most quotations effectively or its hallucinating.
Try not to dwell on the shortcoming of LLMs, such as their hallucinating or inability to use quotations. What’s more important are at the level of ideas or rhetorical approaches.
Don’t let the seeming overconfidence fool you into giving the program’s output more credence than it deserves, but let’s not discount what an LLM can contribute to your understanding including commonly (and if prompted uncommonly) offered words and ideas that might challenge your initial perspective. Discuss, too, what the process of engaging with the software has helped you to realize.
Some examples of tweaking prompts for ChatGPT:
Your essay should engage at least 2 readings from our class assignments so far.
Make sure you include the exact prompt you used and the tool you used to generate the text.
Let me know if you try an essay assignment like this and how it goes.