Using Temporary Files to Simulate Internal States in ChatGPT

ChatGPT Use cases and examples
1

ChatGPT lacks the ability to internally store and iteratively refine drafts. To simulate this, we used temporary files as a simulated internal buffer, allowing for the creation, storage, and refinement of drafts.

We used the following custom instructions:

Draft Creation and Storage:

  • Generate an initial draft.
  • Store this draft in a temporary file.
    Reading and Refining:
  • Read the draft from the file.
  • Refine the draft with additional details and improvements.
  • Present only the final refined version, providing a download link to the original draft.

What we learned:
Using temporary files to simulate internal states in ChatGPT improves output quality.
This can also be used for planning rather than drafting.
The main issue is that these responses take much more time as chatgpt must store and retrieve the temporary file.

So interesting, but mostly useless.

Are there any alternative methods that actually work? What are your thoughts and theories?

2

We’ve also tested using the memorisation feature for this same purpose. It seems to work and is much faster than using temporary files. Still experimenting whether either method is truly effective.

3

IMG_2744
IMG_27441170×2532 275 KB

IMG_2745
IMG_27451170×2532 128 KB

IMG_2746
IMG_27461170×2532 147 KB

These screenshots show the responses and how they differ using this process when asked a single question:
"Very consisely describe the process of water freezing "

They also show the contents of the draft document.

This shows that this method does improve answer quality at expense of speed.