Operationalizing Foundation Models: The Next Frontier for Enterprise AI

The speaker discusses the importance of Foundation models in AI and the human-centered design approach at IBM research.

• Foundation models are already impacting human beings and will likely power more than one third of AI in the Enterprise by 2025
• There are different types of Foundation models, including general purpose, domain-specific, and enterprise-specific models
• Human-centered design is a central principle at IBM research and involves considering the different personas that will interact with AI systems
• Efficiency and resource efficiency are important areas of innovation in the creation of large models
• Enhancing representations with reasoning capabilities is a future goal for AI

The speaker mentions that Foundation models have already happened in NLP and are not far out, which is important for business leaders to consider. They also discuss the importance of considering different personas when designing AI systems, such as data scientists, process engineers, and business leaders. The speaker also highlights the need for efficiency and resource efficiency in creating large models, as well as the future goal of enhancing representations with reasoning capabilities.

Sriram Raghavan, Vice President of IBM Research for AI, discusses foundation models with Vijay Karunamurthy, Scale’s Director of Engineering. Learn how Foundation Models allow enterprises to leverage recent research, while also fine tuning to their respective use cases and (often smaller) training datasets. Raghavan will explain how the effectiveness of Foundation Models comes from their ability to generalize nearly any data type and domain as another language. This is the same flexibility that helped IBM Watson double its number of supported languages in a year. Since then, as Raghavan will cover, IBM has demonstrated that it’s surprisingly effective to extend foundation models to time series and tabular data, simply by treating them as new forms of language. He also discusses how to bring trust into the training process when working with transformers, and the most effective way to employ synthetic data for specific use cases. Prior to his current role, Raghavan was the Director of the IBM Research Lab in India and the CTO for IBM in India/South Asia. He began his career in IBM at the Almaden Research Center in San Jose, California where he led a number of research efforts at the intersection of natural language processing, data management, business analytics, and distributed systems.