Discussion about this post

User's avatar
Shoni's avatar

This is so fun! We did something similar when I challenged ChatGPT to a story battle and had a friend wrangle something out of it using the same plot (you can read her experience with it here: https://shonistar.substack.com/p/behind-the-scenes Scroll down to Heidi's section). I feel like you got better results, maybe because it was an earlier version, or maybe thanks to your brutally honest feedback, but the style of the stories is actually quite similar, and the titles!

Why wouldn't you give feedback like that to a student, out of interest? Too personal? Do you think it would have the opposite effect on a human than it did on a machine? Maybe put them off writing and iterating? That's interesting in itself.

Expand full comment
Marcella Gemelli's avatar

I always appreciate hearing about ChatGPT experiments and the thoughtful analysis of process and outcomes. I have to say that I was maybe looking for a part 3 that would discuss more in depth the question posed about authorship and perhaps some other issues with ChatGPT as novelist. Firstly, as you noted, how do we sort out the authorship/editor roles? This concerns me as we have already seen students submitting ChatGPT's work as their own but also in how we may possibly completely flip roles where ChatGPT is the creator and we as humans only serve as editors. How does this play out in the future of master's theses and PhD dissertations? Which brings me to a second related concern - a standardization of writing. The comment below about a different person providing a different prompt with a result of stories and titles being similar begs a question about the limits to ChatGPT's creativity and style. A third concern is trust. Perhaps a student is tasked with a writing assignment and asks ChatGPT to write them a 5,000 word creative story in the style of John Wyndham. The student didn't actually read the assigned novel by Wyndham and so cannot adequately determine whether the style is correct. The student submits ChatGPTs creation without evaluation. I find that we are putting a lot of emphasis on experts being able to drive and evaluate ChatGPT responses but we may not always have that capacity. Thank you.

Expand full comment
3 more comments...

No posts