Options for utilizing generative AI in business
When using generative AI in practice, there are several options.
RAG (Retrieval-Augmented Generation)
Objective: To expand information by imparting external knowledge to LLM.
Cost and time: relatively cheap and fast.
Difficulty level: Simple because it gives external information to large-scale language models such as ChatGPT.
FineTuning
Objective: To be able to do an LLM with specialized knowledge.
Cost and time: Relatively expensive and takes longer than RAG.
Difficulty level: Since LLMs such as llama3 are subject to additional learning, it takes less man-hours than constructing a large-scale language model.
Large Language Models
Objective: To create a new LLM using only in-house learning data.
Cost and time: Expensive and takes longer than RAG or fine tuning.
Difficulty level: Annotation effort etc. are required to build a language model in full scratch.
