GenAI risks and global adoption - Practical AI Recap

Podcast: Practical AI

Published: 2025-08-27

Duration: 43 min

Guests: Rick Kobayashi, Kenny Song

Summary

The episode examines the global adoption and risks of generative AI, highlighting Japan's unique position in embracing AI technologies. Citadel AI's founders discuss the importance of testing and monitoring AI systems to ensure safety and reliability.

What Happened

In this episode, Rick Kobayashi and Kenny Song from Citadel AI discuss the adoption of generative AI in Japan, noting that while Japan lags in developing foundational models, it excels in the application of GenAI technologies. This is attributed to cultural factors, such as familiarity with AI through popular media like anime. They explore the regulatory landscape in Japan, which is between the US's lax approach and the EU's strict regulations, focusing more on guidelines for reputational risk rather than strict legal requirements.

The conversation dives into the specific challenges that enterprises face when deploying AI systems, particularly in Japan's financial sector. These companies are eager to implement AI to improve services but must tread carefully to avoid reputational damage or regulatory repercussions. The importance of balancing innovation with risk management is emphasized, especially in heavily regulated industries.

Kenny Song explains Citadel AI's approach to AI safety, focusing on testing and monitoring AI systems to mitigate risks such as hallucinations, toxicity, and prompt injections. He describes how Citadel's tools help organizations transition AI models from proof of concept to production-ready systems by offering robust testing and real-time monitoring solutions.

The discussion also covers the differences between startups and large enterprises in adopting AI. Startups, with fewer brand risks, can experiment more freely, while larger companies require stringent testing and monitoring due to mature risk management practices. This reflects the need for custom solutions tailored to a company's size and risk appetite.

Rick and Kenny highlight the evolution of Citadel AI from its inception in 2020, initially focusing on traditional predictive models, to now addressing the complexities of LLMs. They emphasize the critical role of human judgment in setting evaluation standards, using LLMs as tools to automate and scale these evaluations.

The episode concludes with insights into the future of AI, with both guests expressing optimism about the rapid technological advancements and the opportunities they present. They discuss how Citadel AI is positioned to help businesses navigate this evolving landscape by ensuring AI systems are reliable and trustworthy.

Key Insights