from Diary 2023-07-18
Integration of LLM into business systems and its potential
Summary: Although it is difficult to incorporate LLM into existing business systems, the concept of systemizing parts that could not be systemized before with LLM is necessary, and a bottom-up approach to the software development process is also required.
@kis: in terms of how to integrate LLM into existing business systems, the conclusion so far is that LLM has no use in existing business systems. I feel like. It could probably be used supportively in places, though. Existing business systems are a logical and organized part of the business.
-
@nishio: the existing “electronic systems” have no use, and the existing (left over because they can’t be electronic) “human systems” are going to be LLM. I think that the existing “electronic system” has no use. So looking at the current system and thinking about where to change it is narrow-minded at that point.
@kis: existing business systems are hard to use LLM because you can do it logistically. I think we need to think in terms of “systematizing operations that could not be systematized before through LLM. @kis: But then, I feel that the existing software development process of “finding routine parts of a business and structuring them to create a system” also needs to be reviewed. I think it is necessary to rethink the existing software development process. Maybe it’s not a top-down approach, like hearing and structuring, but a bottom-up approach, like taking actual data and converting it to LLM where possible, or something like that. @tokoroten: @kis I drew this diagram a while ago I am writing an article about the transition of the way business is done based on this diagram, and it will appear in a certain place in the next month or so.
@kis: @tokoroten That’s about right. I think that the main goal is to use LLM to format irregular inputs for processes where the input is irregular and the output is regular, like Function Calling.
This page is auto-translated from /nishio/LLMの業務システムへの組み込みとその可能性 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.