Purpose of Activity

Regarding AI and LLM, which continue to spread and expand, we will examine and clarify the feasibility of providing Japan-specific multimodal AI services by switching between multiple LLMs, including not only models developed by BICTEC but also domestically produced LLMs that have an advantage in Japanese language processing, depending on the use case. It is assumed that LLMs will be implemented in the cloud and on-device, and that they will be connected and integrated using information and communication technologies and infrastructure such as 5G and IOWN to perform optimal processing using information from different sources.

Activities

A desk study will be conducted on the architecture to realize multi-modal generation AI services supporting multiple LLMs. To this end, discussions will be held with LLM vendors, cloud vendors, device vendors, telecommunication carriers, etc., and a detailed study of the technical means to realize this service will be conducted.

[XGMF-PJ-2409]Feasibility Study of a Multimodal Generative AI Service Using Multiple LLMs with ICT (XG for LLM) [2024.10.21]

project leader

NTT DOCOMO, INC.

OKANO, Yoshiki

subleader

NTT DOCOMO, INC.

MORIBE, Tomoya