image

  • I slept a lot.

🤖2023-08-13 07:08

  • I wrote in my last note about my considerations developed in Co-operation with AI and they were ignored in their entirety.
  • After all, if you have a feeling that you want people to use this, you should explicitly instruct them to do so.
  • If you want to control it properly, you should separate the steps.
  • The previous note should not be repeated in its entirety.
  • It’s not a good idea to throw away all the notes from last time.
  • It’s important that the thoughts develop from the previous note.
    • Otherwise, we’re just doing one random gacha per day, and there’s no point in having a “previous note”.
  • Better to find a connection between the previous note and the current fragment

from the start

Read the information provided, which consists of random fragments from a colleague’s research notes You can read your own research notes This prompt clearly tells me that random notes are more important than my own notes, my English is bad.

Read your previous research notes, which are essential You may also read the random fragments from a colleague Nishio’s research notes Changed it to this.

AI can write diaries too, because humans write diaries.

Fine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Models to Unique Applications

@overlast: The current status of LLM service application is “Of course we need a high performance inference API, but there are a lot of functions that need to be developed to “adjust the input/output according to the law”. But there are a lot of functions that need to be developed to “adjust the input/output to meet the legal requirements”, and it takes 2-3 years each to build them to have high versatility and excellent UX”. I feel like I’m watching coffee boil on a coffee server. ☕️


This page is auto-translated from /nishio/日記2023-08-13 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.