Interview, Fireside Chat
RAISE 2025: Building AI for Security-Minded Enterprises
- Mathematical PhD backgrounds in AI are expected to drive more applied and concrete impacts on daily life.
- The organization plans to continue training foundational models to achieve tangible business results.
- Investment is directed toward on-premises hosting capabilities and customizable model controllability through co-development partnerships.
- Significant resources will be dedicated to providing flexible, tightly secured agentic capabilities.
- Security measures embedded within models cannot guarantee absolute safety due to the stochastic nature of these systems.
- Grounded generation will be enhanced via retrieval-augmented pipelines, web search, and agentic behaviors to increase trust in highly regulated sectors.
- Trust in model outputs relies on the reliability of source documentation rather than a specific quantitative guarantee from the model alone.
- AI models are predicted to solve problems through methods distinct from exact human-style reasoning.
- Co-development modeling partnerships will be replicated to create a Korean language model in collaboration with LG, following a similar approach used with Fujitsu for Japanese.
- Language capabilities will be developed using a global mix of data and partner domain expertise instead of relying solely on fine-tuning.
- Synthetic data will be utilized to augment existing datasets, fill gaps, and improve output verifiability through ingrained verifiable rewards.
- Strategic focus is set on creating value and building secure AI for corporate clients rather than pursuing Artificial General Intelligence (AGI).