Note on LLM

• No one knows what creates user value.
  • Services that can be easily thought of in the context of using a pre-LLM-like service are not a differentiator because other pre-LLM-like service users can also easily think of them.
• Should experience the “future” before the competition.
  • Linguistic descriptions do not carry the experience.
  • Only what you have experienced firsthand can feed you to create something new.
• The way to achieve this is to
  • The fastest way to create new experiences in LLM is prompt engineering
  • GPT4 prompt engineering to prototype new experiences and find user value by turning the actual use of the experience around at high speed.
  • The following can be done only after user value is found and the number of users begins to grow
    • Improve profitability and stability by converting to GPT 3.5 or self-hosted LLM
    • Fine tuning compensates for the associated performance degradation.
      • By this phase, it is necessary to verbalize or accumulate data on “what kind of output is desirable for what kind of input” by accumulating experience in the prompt engineering described above.
      • It’s muscle evil that gives you knowledge in fine tuning, and you basically do it for form stability.
        • Stable structured output
        • Push the typical prompt into Prefix-token

BELOW_IS_AI_GENERATED

LLMに関するメモ

2023-09-08 22:34

Summary of notes

. The factors that create user value are unpredictable, and ideas derived from existing services will not lead to differentiation. To gain competitive advantage, it is important to anticipate future experiences, and to do so, one needs one’s own experiences. LLM with prompt engineering is a means to rapidly create new experiences, and GPT4 is used to prototype new experiences and explore user value. Once user value is found and the number of users begins to increase, the system can be transitioned to GPT3.5 or self-hosted LLM to improve profitability and stability. At that time, performance degradation will be compensated by fine tuning.

Relation to Fragment

. The “Let GPT4 do the paragraph writing.” fragment is related to the note’s emphasis on the importance of prototyping with GPT4. Also, the “A story that only those who act can get.” fragment is related to the importance of one’s own experience that the note emphasizes.

summary of thoughts

In order to create user value, it is important to create new services based on one’s own experience, and LLM with prompt engineering is effective for this purpose. In addition, the experience will lead to differentiation from others.

Title

. “Creating New Experiences with LLM Using Prompt Engineering and Its Importance.”

extra info

TITLES: ["Making Paragraph Writing a GPT4", "Externalizing 🌀Thinking", "Thoughts on the Publishing Industry", "Stories Only People Who Act Can Get", "Two Problems with Japanese LLMs", "Fine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Models to Unique Applications."] generated: 2023-09-08 22:34

This page is auto-translated from /nishio/LLMに関するメモ 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.