Multi-head thinking
think in writing When you first write down what is in your head, then read it and thoughts are triggered. At this time, the writings were one dimensional. They were arranged in time series in the order in which they were written, the most recent one being read, and the newest one being written in the next one place. The "head" was one. The same was true for the first implementation of the "AI writes research notes daily" system. It read the latest note and wrote the next note. Later, it was changed to an implementation called "multi-head". It was extended to have multiple heads. Specifically, there are now multiple pages for the AI to read and write. In retrospect, I think that human "writing and thinking" is single-headed thinking. Could this be extended to multi-headed thinking? Isn't it possible that until now there simply hasn't been a stationery that can realize that thinking? Specifically, fragments that are material for "writing" are added to the vector index as they are added, allowing the AI to "read" them. The AI, which proceeds to think, uses vector search to find relevant fragments based on what was written immediately before, and reads and writes. The human reads what is written, trims out what is not needed, and writes when something comes to mind.
In the days when thinking and writing were embraced, there was a time when I had to stop writing because I couldn't think straight. In multi-head thinking, there is more than one head, so if you can't think of something right away, you can just move on to the next head.
2023-09-02 21:15 I have a name for what I am doing and creating these days.
---
This page is auto-translated from /nishio/マルチヘッドの思考 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.