Journaling with GPT

I have been tinkering with 'tools for thoughts' for years.

The problems they suffer are well documented by https://every.to/@danshipper

I have also been experimenting with chatGPT, and am astonished by the results.

Dan brought my attention to a github project named gpt_index which solves the problem of chatGPTs short memory.

Currently, chatGPT only allows 2,048 tokens (approximately ~1,500 words).

Basically gpt_index allows you to provide a folder of many entries (journals, essays, etc), and feed it recursively to a LLM (large language model). gpt_index will then store the results in a data structure. Then, use that data structure to query the LLM.

I input some journals over the years and the results were unsurprisingly remarkable.

It did take some promp tweaking to avoid default answers. Using syntax like "what are the authors goals" or "write a biography for the author".

This particular project uses python, which I don't get to use often, so that was enjoyable.

My name isn't John, I'm only 30, and I haven't worked in video and motion design for years – but nonetheless, much potential!