Beyond the basics
As we move into the second year of living with artificial intelligence as co-creator, it sometimes becomes necessary to bring clarification to misunderstandings surrounding the development of an application like Subtxt.
Over on the Discuss Dramatica forums (a site that I admin focused solely on the Dramatica theory of story and the Dramatica application), a brand new visitor to the site left an interesting comment:
Although Subtext [sic] is an excellent tool, it is ridiculously overpriced. Especially now that ChatGPT can offer similar writing assistance (plus LOADS of other features) for $5 less than Jim Hull’s app.
All one has to do is go to “Create my GPT” and upload the free Dramatica ebooks with proper instructions.
I’ll get to the pricing in a second, but first this idea that “all you have to do” is upload PDFs and eBooks from Chris and Melanie to get the same response.
Actually, no--I’ll let someone else who has tried that--explain the results:
I did try with OpenAI. I also imported books and analyses and so on. My advantage here was I knew what’s in these books. And after trying with Dramatica and some other topics I think, please don’t hate me for that, I think it’s kind of ‘stupid’. Trying to test it sometimes I had to change the way a question was asked like ten times to get the correct answer.
There is a tremendous amount of work I do behind the scenes to make Subtxt and Subtxt Muse as intelligent and as helpful a story development chatbot can be in 2023. This work goes above and beyond just uploading PDFs and eBooks and requires much more effort than simple prompt-engineering.
Greg Brockton, OpenAI’s top engineer, recently said “evals are all you need.” This is a take on the famous AI paper “Attention is all you need” (which is what basically led to ChatGPT and other amazing advances in the field). In short, the idea is that--above all else--being able to systematically evaluate the interaction between user and AI is what leads to it growing more and more intelligent.
Working with Dramatica + ChatGPT within just ChatGPT is working with theory in isolation--without the added intervention of experts--and without the emergent properties that arise from RLHF efforts (real-life human feedback). These constant evaluations of both interaction/response and the accuracy of said interaction/response makes all the difference.
As another user pointed out:
I’m a long time user of Subtxt from way before any of the chat stuff existed. For me, there is still a huge amount of value in everything else the app provides, like quick access to wonderful pages on each element (showing relevant quads, example illustrations, movie references for that element, etc.).
All that “stuff” before chat existed is just another part of the evaluation process. All those illustrations, movie references, analyses, podcasts, support questions, articles, and classes? Those are all a part of the Subtxt experience.
In regard to pricing, I mention it in the forum post, but don’t mind repeating it here:
I understand that not everyone can afford a subscription and have in the past given out temporary free licenses to those who are under financial distress or who are in parts of the world where the fee is beyond their means (purchase parity).
All you have to do is email me (or contact me here through DM, though Discord might be quicker) and I’ll see what I can do.
Don't miss out on the latest in narrative theory and storytelling with artificial intelligence. Subscribe to the Narrative First newsletter below and receive a link to download the 20-page e-book, Never Trust a Hero.