Re: Blaugust and Large Language Models
Belghast wrote about Blaugust and Large Language Models. I started to write a comment. It ended up longer than excepted and decided to make a blog post instead.
I use iA Writer and the developer has written about their opinions on using AI. They refused to implement AI integration to the application and instead added a feature to keep track which parts of the text originate from AI. Their suggested approach is to use AI a sparring partner not the writer. Instead of prompting the AI to write something for you make it ask you questions. Then use your answers as the basis of the text.
I have not tried this approach yet, but find it interesting. Last Blaugust I tried to make a text adventure with ChatGPT. The output was bad and I quickly gave up the project. With the above approach I might have at least come up with a decent outline of the adventure.
I have been pondering parallels between AI generated images and photography, which was controversial when invented. Photographer frames the subject, presses a button and gets an image. With AI one makes a prompt and gets an image. Analogue photographs had to be developed but in the digital era post-processing of the both images is similar. The above example is simplified, but raises the question at which point an endeavour becomes creative?
Is it about the how accepted the method is, how complex the process or how much input is needed from the artist? With AI you could spit out a prompt and pick the first image which look nice. Alternatively, you could plan the desired output, experiment with different prompts, do in/out-painting, and combine the results with image prompts. Are both approach equally uncreative? I like to think the latter approach is at least somewhat creative, but letting AI do all the work without any thinking is undesired.
To illustrate the above point, below is a photograph of a daisy and AI rendered cat playing synthesizer. I did not create the subject in either of images and they required approximately equal effort. I chose the desired composition and applied post processing. Is there a difference in the perceived creativity of the two images?
For me the largest problem is not the use of AI itself, but using it to spam large quantities of junk content or corporations scraping massive amounts data without permission or compensation to train their models.
A photograph of a daisy. The daisy grew itself. Photo composed and post-processed by the author.
AI image of cat playing syntheziser rendered by Midjourney. Prompting and post processing by the author.