from Diary 2023-09-15 Consideration of the long sleeping oracle
This time the sleeping oracle is long, there are three cases. They appear to be independent, but some are related. Do you want to give AI Shaman both the whole thing and the individual things to try?
Longer ones are strongly “summarized” and not interesting.
- The fixed width of the LLM makes it impossible to inflate it with longer inputs, making the summary stronger
- Reasonably sized for the processing you are expecting.
- This is implementation-dependent and can change with other LLMs, and the same LLM can be affected by the prompt.
- The length of the input probably influences the behavior of the LLM as a nonverbal instruction
- There may be a moderate size factor on the hardware side, the human factor.
- Neither a phone twice as big nor twice as small is a good phone to use.
- People have their own preferences for the thickness of the pen barrel.
- As far as AI shamans are concerned, this is like a milestone for [go
- Instead of lining them up tightly as in A, loosely enclose them as in B1.
- The perimeter will be supplemented later, as in B2.
- In the case of Go, they add stones, and in the case of AI shamans, the LLMs fill in the words.
- For that to happen, there has to be room for more stones than the first ones placed, because there will be more stones placed than the first ones
You’ve let the AI develop is being dragged down by the “AI shaman” in the opening commentary.
- I had to cut it down and start over.
- Answers Behind the Clouds
- Mistakes were made and some were made from the title only. - Behind the Clouds Answer 2
This page is auto-translated from /nishio/長い寝起き神託の考察 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.