I teach a university class in which I ask the students to submit writing each week, and I have also seen obviously LLM-produced writing. Yes, it’s boring and doesn’t show the students’ thinking, and the students are not getting any wiser by doing assignments that way. Just last week, I told my students that, while they can use LLMs any way they like for the class, their writing will be more interesting if they write it themselves and use LLMs only sparingly, such as for fixing grammatical mistakes (most of the students are not native speakers of English). It helps, I think, that in this class the students’ writing is shared among the students, and during class I often refer to interesting comments from student writing. The students themselves, I hope, will come to understand the value of reading human-written writing.
That said, I myself am increasingly reading long texts written by LLMs and learning from them. I have been comparing the output of the Deep Research products from various companies, often prompting for topics that I want to understand more deeply for projects I am working on. I have found those reports very helpful for deepening my knowledge and understanding and for enabling me to make better decisions about how to move forward with my projects.
I tested Gemini and ChatGPT on “utilizing Euler angles for rotation representation,” the example topic used by the author in the linked article. I first ran the following metaprompt through Claude:
Please prepare a prompt that I can give to a reasoning LLM that has web search and “deep research” capability. The prompt should be to ask for a report of the type mentioned by the sample “student paper” given at the beginning of the following blog post: https://claytonwramsey.com/blog/prompt/ Your prompt should ask for a tightly written and incisive report with complete and accurate references. When preparing the prompt, also refer to the following discussion about the above blog post on Hacker News: https://news.ycombinator.com/item?id=43888803
I put the the full prompt written by Claude at the end of the Gemini report, which has some LaTex display issues that I couldn’t get it to fix:
https://docs.google.com/document/d/1sqpeLY4TWD8L4jDSloeH45AI...
Here is the ChatGPT report:
https://chatgpt.com/share/681816ff-2048-8011-8e0f-d8cbad2520...
I know nothing about this topic, so I cannot evaluate the accuracy or appropriateness of the above reports. But when I have had these two Deep Research models produce similar reports on topics I understand better, they have indeed deepened my understanding and, I hope, made me a bit wiser.
The challenge for higher education is trying to decide when to stick to the traditional methods of teaching—in this case, having the students learn through the process of writing on their own—and when to use these powerful new AI tools to promote learning in other ways.