For the experiments to be meaningful, I need to compare several approaches in each experiment, which requires the creation of several files:
- Basic inputs
- Summariser 0 is the original Summariser, written off the cuff for the ChatGPT pilot without much thought. It creates very short summaries and so serves as our "baseline Summariser" (see postscript).
- C-1_s-0 was created using Summariser 0 (ie short summaries of the Collection 1 notes)
- hence c-1_s-0-response-p-1: what ChatGPT returns when sent C-1_s-0 and asked Prompt 1
- Summariser 1 (see summarisers in summary#^29496d)
- AllNotes: to test if and how Summarisers affect the quality of ChatGPT's response, we need to apply the Prompt to the entirety of the notes, as well:
- finally, the analysis: Experiment 1 analyses, itself written with the help of ChatGPT (experiment 1 analysis chatgpt execsummary).