Provides an executive summary, list of salient details, and some editorial writing feedback for any chapter up to about 6000 words or so.
- Add your
key_openai.txt
file - Create folders
chapters_doc
,chapters_txt
andsummaries
- Copy your Word docs (DOCX) into the first folder. One chapter per file
- Run
step01_convert_docx.ps1
you may need to have Word/Office installed - Run
step02_summarize.py
you will need OpenAI installed and access to the GPT4 API
This will generate a short report for each chapter. You can use the summarizations to feed into ChatGPT or other language models. You can also use the feedback to guide your editing journey.
For the developmental feedback, I found that asking the model "What's the one biggest problem?" tends to yield the best results. Since editing is an iterative process, it makes sense to focus on the biggest problem, one at a time, and incrementally improve your work. This is beneficial because human brains can only focus on one major task at a time anyways. Too much feedback would be overwhelming.
Keep in mind that the model is looking at your chapter in isolation and so it may miss the point. Take it with a grain of salt.
For the prose feedback, it will focus on the line level, providing some general advice about your writing style. It is geared towards a modern writing style, which should include a good compositional balance, including action, dialog, interiority, and description. This is called "prose layering". As with developmental feedback, take it with a grain of salt. Take what makes sense and leave the rest.
I created this based on my work with a professional editor. They provided a chapter-by-chapter report that was far more detailed than this model's output. A human editor, that you can have a conversation with, is still superior to GPT4, but maybe not for long. Especially as token windows get larger, and you can have a hypothetically unlimited conversation about your entire chapter (or novel!) editors may have a hard time keeping up.
With all that said, there are still elements of the human experience that the machine does not yet (and may never) grasp. For instance, the subjective feeling of resonance when you read something that connects emotionally, something that feels moving and emotionally true. These models are still expensive (relatively speaking), as I spent $8.02 reviewing the first 19 chapters of my novel. This is still far cheaper and faster than a human editor.
To use this, you'll need to do the following:
- Populate
scratchpad.txt
with whatever you wish, up to 6000 tokens or so. You can copy thesummaries
intoscratchpad.txt
, which will then be used by theSYSTEM
field of ChatGPT API, which means the model will be instantly aware of your story. - To use the TKINTER chat interface, just use
chat_tkinter2.py
. This will open a small, simple window. UseSHIFT+ENTER
for newlines in the chat window. It will also be color coded. Notice the "MUSE is thinking..." output in the bottom right while the API is generating a response.