In attempting to create a script through ChatGPT 3.5 in the previous midterm project, my results were fairly inconclusive on the creative aptitude of the artificial intelligence’s software. To clarity this statement, while the program was capable of taking a single director’s name and form at least a three-page script output, the inventiveness and rhetorical capabilities of ChatGPT 3.5 are still fundamentally formulaic. The three questions demanded of ChatGPT 3.5 concerned first a broad genre of filmmaking, a romantic comedy, a more specified genre, a Roman epic, and finally a single director as the writer, PTA (Paul Thomas Anderson). All three of these attempts resulted in a similarly formulaic writing output, while the rom-com was set in a NYC café with a barista protagonists, the Roman epic concerns Marcus Aurelius, author of Meditations. Finally, it is in choosing the director’s name, Paul Thomas Anderson, that this project attempts to redefine the writing methods of ChatGPT to inherently make itself creative and less formulaic; this did not succeed here.
The biggest issue seemingly stemmed from the script devised from the PTA input being so strongly a copy of his most recent work, Licorice Pizza. Fundamentally, well-known directors, particularly those who are still producing pictures today, will inherentely derive ChatGPT to create a script based on the most popular result i.e., their most famous or most recent work. Thus when attempting to use an input for Steven Spielberg or Stanley Kubrick, the scripts, “Search for Atlantis” and “Temporal Anamoly Unveiled”, never stray far from the most well-known and popular aspects of each director’s work. Thus the solution may plausibly be to include director’s that are not extremely well-known, and thus implicate ChatGPT 3.5 as a creator of work that does not have as much inherent content to base itself on.
I have attempted to use this deep-dive into film director’s to force ChatGPT to make inherently creative work. I have begun by including directors like Fernando Di Leo, Andrej Zulawski, and Djibril Drop Membéty, to hope to force ChatGPT into a position where it cannot be forumlaic in its writing. Nevertheless the Zulawski script derives itself from his most famous work Possession (1981), while the Di Leo script shows some promise in less formualic work, as the dialogue and plot though clearly based in Di Leo’s work, is not nearly as parasitic in its nature as an ‘original’ script.
Leave a Reply