A casual replication of part of "Why Literary Time is Measured in Minutes" using ChatGPT and GPT-4 through the OpenAI API.
Please note this is casual work for a blog post, and is currently documented lightly. Also a couple parts are not done as carefully as I would for an article: e.g. in the bag of words solution I select features for test and training at once, pooling both sets, which is not good practice if you're being careful.
The API interactions are in GetTimes.ipynb (which uses Chat) and GetTimesGPT4.ipynb (GPT-4).
The output of those scripts goes to output2.tsv (for Chat) and outputGPT4.tsv (for GPT-4). There you can see the original passages of fiction, plus the Assistant's responses to them, and times estimated by both the human reader and the model.
I'll document this a little more carefully later today.