
2024/10/30
On Tuesday, July 16, 2024, GENIAC hosted its second expert event in a hybrid format, both online and offline. Two specialists from Microsoft, experts in AI foundation model development, gave presentations on the latest trends in AI model development, societal implementation, and future outlooks. They shared insights with project participants and stakeholders, fostering stronger community connections.
Microsoft’s SLM "Phi-3" Development Explained
The first online speaker was Hardik Modi, Principal Product Manager leading the development of Microsoft's SLM (Small Language Model) "Phi-3". The presentation was delivered in English, with simultaneous interpretation available, followed by a Q&A session with the participants.
Mr. Modi explained the vision behind AI development and highlighted the ongoing development of the "Phi-3" family. He emphasized that although these models are smaller and more lightweight compared to traditional LLMs (Large Language Models), they offer a unique balance of cost-efficiency and high-quality performance. He concluded by pointing out that even smaller models can achieve high performance in specialized domains, such as healthcare, making it possible to build AI models optimized for specific use cases.
Current Security Measures for LLM Applications and Examples of Societal Implementation
The next speaker was Nobukatsu Tanahashi, an AI/ML Specialist from Asia Global Black Belt. His presentation, titled "LLM App Development and Societal Implementation," focused on the security of Phi-3 series models, content filtering mechanisms, and specific AI application use cases.
Mr. Tanahashi discussed how ensuring safety involves identifying vulnerabilities, continuously improving the models, and performing vulnerability testing for security and AI risks. This enables companies to address specific categories of risk.
He also highlighted the importance of three key phases when introducing LLMs into companies: empowering employees, improving operational efficiency through internal DX (Digital Transformation), and enhancing customer engagement. Each phase plays a vital role in the successful deployment of use cases.
During the Q&A session, participants raised questions about how Microsoft builds domain-specific expertise and highly specialized datasets.

Looking ahead, GENIAC will continue to organize events and study sessions with AI development experts, as well as matchmaking events between selected project participants and user companies. Please stay tuned for GENIAC's future activities as they continue to lead AI utilization in Japan.
GENIAC Top PageLast updated:2024-02-01