I wanted to share with you guys a little program outline that theoretically would give every message insane memory recall across conversations with your AI. This program centers around the idea of generating weighted memories and recalling them with context windows.

The outline is admittedly extremely simplistic, but the idea I believe holds merit. If someone were to streamline this concept, and build it into an LLM’s structure I would be interested to see how it performs.

Let's make a python program with a visual interface.

Basic functions: Create a project to contain the work about to be done. Use the ChatGPT API to allow the user converse and get responses. Use ChatGPT 4 for main user input/output. Use ChatGPT 3 for getting keywords, and doing general searches. Adjust custom instructions to ChatGPT. Store selected paragraphs, or entire parts of the conversation to a text file.

Work Flow: The user creates a new project and names it. The user creates a new chat within that project and names it. The user will query ChatGPT 4. Once the response is received, the user can select part, or all of the response and select "Remember That". Once something has been remembered, the "Memory Assist" toggle becomes available. They can then continue to query.

Back End: When the user creates a project the program creates a new folder with the project name. When the user creates a new chat the program creates a new folder within that project folder with that chat name. When a user selects parts of the response, the program will take that selection and send it to ChatGPT 3. It will ask ChatGPT 3 to go through the text and find up to 10 keywords, and derive a title for the text. The program will also ask ChatGPT 3 to assign those keywords a weight from 1-10, based on how important it is to the selected text. The program will then save the text to a text file, with the title given by ChatGPT shortened to 10 characters, and the date, sequentially numbered if there is a duplicate, as the file name. Then it will add a record to a project XML file in the base project folder. This record will contain the text file path, the full title given by ChatGPT, the keywords and their associated weights as one object or record in the XML file.

IF "Memory Assist" is on: When the user sends something to chat gpt, the program will first use the user input to search the XML file. It will look in each record, and find any keywords in that entry that associate with the user input. Then it will total up the weights of those keywords to give that record a total compatibility score. Then it will pick the two highest compatibility scores.

Once the two records have been selected, it will attach the two text files associated with those entries to the user's query to ChatGPT. Thereby giving the input the context window for it to respond accurately.