GenAI Superstream: Developing Innovative Apps with Generative AI

Chloé Messdaghi: Introduction (5 minutes) - 8:00am PT | 11:00am ET | 3:00pm UTC/GMT

• Chloé Messdaghi welcomes you to the GenAI Superstream.

Rebecca Gorman–Keynote: The Commercial Value of Building Ethics and Safety into GenAI Apps (15 minutes) - 8:05am PT | 11:05am ET | 3:05pm UTC/GMT

• Generative AI is enabling the development of innovative applications and solutions, but alongside the potential of this powerful technology lies the responsibility to consider the ethical and safety implications for all stakeholders. Rebecca Gorman discusses some of the most crucial considerations for responsible development with generative AI and the risks businesses face by ignoring them. You’ll also hear how integrating ethical and safety measures not only mitigates risks but can also produce significant commercial value.
• Rebecca Gorman is a serial entrepreneur, seasoned technologist, and AI expert. She built her first AI system 20 years ago and has advocated for responsible AI since 2010. She has co-developed several advanced methods for AI alignment and advised policymakers in the EU, UN, OECD, and UK Parliament on responsible AI regulation. Rebecca was named in REWork’s Top 100 Women Advancing AI in 2023, nominated for VentureBeat's Women in AI Award for Responsibility and Ethics in AI, and is a member of the Fortune Founders Forum.

Michael Running Wolf: Generational Data Sovereignty for Indigenous Communities (30 minutes) - 8:20am PT | 11:20am ET | 3:20pm UTC/GMT

• Indigenous data is underrepresented in large AI datasets, but closing this gap with data collection creates ethical risks. Native American cultural heritage, including sacred documents and oral histories, has been lost or damaged due to exploitative anthropological romanticism and outright theft. Generative AI’s insatiable thirst for data risks the exploitation of Indigenous knowledge. Protecting Indigenous data requires security considerations, intellectual property law, and respect for Indigenous ways of knowing. Michael will discuss practical solutions for collaborative ML research with Indigenous communities.
• Michael Running Wolf, Lakota and Cheyenne, was raised in rural Montana with intermittent water and electricity. He’s a computer scientist, a former engineer for Amazon’s Alexa, an instructor at Northeastern University, and cofounder and lead architect of First Languages AI Reality. Michael uses AI to reclaim Indigenous languages and has been awarded an MIT Solve Fellowship, the Alfred P. Sloan Fellowship, and the Patrick J. McGovern AI for Humanity Prize.

Jay Alammar: Large Language Models as Building Blocks (30 minutes) - 8:50am PT | 11:50am ET | 3:50pm UTC/GMT

• The rise of large language models is inspiring a wide variety of application ideas and experiments. Creators who want to become more adept at building with LLMs must gain an understanding of using LLMs as components of advanced pipelines—not just as a text-in–text-out monolith. Jay Alammar discusses a number of LLM applications, outlining use cases for generative AI and for semantic search and data exploration.
• Jay Alammar is director and engineering fellow at Cohere, a provider of large language models as APIs, where he advises and educates enterprises and the developer community on practical use cases for language models. Through his popular AI/ML blog, he’s helped millions of researchers and engineers visually understand machine learning tools and concepts from the basic (ending up in the documentation of packages like NumPy and pandas) to the cutting-edge (Transformers, BERT, GPT-3, and Stable Diffusion). Jay is also a cocreator of online machine learning and natural language processing courses.
• Break (5 minutes)

Pamela Isom: Navigating GenAI with Cybersecurity and Ethics (30 minutes) - 9:25am PT | 12:25pm ET | 4:25pm UTC/GMT

• Pamela Isom, award-winning AI ethics leader and host of the AI or Not podcast, lays out the opportunities and challenges that GenAI presents, from building trust in AI systems and ensuring robust cybersecurity measures, to addressing the anxiety associated with the rapid pace of technology adoption while upholding ethical standards. Learn how to navigate this exciting frontier and manage the anxiety that can come with disruptive technology.
• Pamela Isom, founder and CEO of IsAdvice & Consulting, has over 25 years of public and private sector experience stewarding safe, innovative digital transformations, corporate governance, and AI and cybersecurity strategies. She was awarded the 2024 Virginia Black Business Leaders Award. In 2023, she was invited to attend the Blacks in Space commemoration ceremony by the National Space Council, and she was recognized as one of the most influential federal government executives. As executive director of AI and technology at the Department of Energy, she led enterprise AI portfolio management, research, and cloud strategy. She also served as deputy chief information officer, chief data officer, and senior agency official for geospatial information.

Suhas Pai: Designing Retrieval-Augmented Generation (RAG) Pipelines (30 minutes) - 9:55am PT | 12:55pm ET | 4:55pm UTC/GMT

• Adoption of LLMs in consumer and enterprise applications has been accelerating, and the retrieval-augmented generation paradigm has played a crucial part in making this happen by facilitating integration of LLMs with external data sources. While current tooling has vastly simplified building RAG prototypes, a production-ready RAG-driven application needs a more sophisticated pipeline, including a lot of software scaffolding. Suhas Pai discusses the stages of RAG pipelines, explores how to assemble them in various ways depending on the use case, and showcases tools and techniques for implementing each pipeline component. He’ll also present a holistic framework for approaching the thorny task of RAG evaluation, highlight the pitfalls and limitations of the paradigm, and explore future research work in this area.
• Suhas Pai is cofounder, CTO, and machine learning research lead at Bedrock AI, a startup operating in the financial domain, where he invented several novel NLP techniques and LM-based architectures that fully power the core features of Bedrock AI’s products. He’s also the cochair of the Privacy Working Group for the BLOOM language model project at BigScience. Suhas is active in the ML community, chairing the Toronto Machine Learning Summit (TMLS) in 2022 and the TMLS NLP conference in 2021 and 2022. He’s also the NLP lead at Aggregate Intellect, an ML community research organization.
• Break (5 minutes)

Michelle Wallig: Leveraging AI and the Power Platform for Active Shooter Mitigation (30 minutes) - 10:30am PT | 1:30pm ET | 5:30pm UTC/GMT

• Addressing active shooter incidents presents physical security teams with complex challenges, calling for a sophisticated response that balances informed decision-making and speed. Michelle Wallig explores how the latest AI technologies, combined with the Microsoft Power Platform, can play a vital role in coordinating and expediting active shooter responses, potentially saving lives. She explains how the Power Platform’s robust automation capabilities can enhance the coordination that ensures a rapid, proactive response to these devastating incidents.
• Michelle Wallig is a 20+ year veteran of Microsoft in four of its IT organizations and two product groups, as well as in sales and services. An AI innovator, she previously worked on neural networks for fraud detection and credit scoring applications in the credit card banking industry.

Sai Kumar Arava: Improving LLM-Based Marketing Analytics Copilots with Semantic Search and Fine-Tuning (30 minutes) - 11:00am PT | 2:00pm ET | 6:00pm UTC/GMT

• Artificial intelligence is widely deployed to solve problems related to marketing attribution and budget optimization. However, AI models can be quite complex, and it can be difficult to understand model workings and insights without extensive implementation teams. In principle, recently developed large language models, like GPT-4, can be deployed to provide marketing insights, reducing the time and effort required to make critical decisions. In practice, there are substantial challenges that need to be overcome to reliably use such models. Sai Kumar Arava shows how to apply a combination of semantic search, prompt engineering, and fine-tuning to dramatically improve the ability of LLMs to execute these tasks accurately. You’ll explore embedding methods, discover how proprietary models, like GPT-4, and open source models, like Llama-2-70b, compare for use cases specific to marketing mix modeling and attribution, and more.
• Sai Kumar Arava is the machine learning manager at Adobe, where he oversees the development and deployment of ML services across various Adobe products. He’s built machine learning solutions that leverage generative AI to solve problems around domain-specific question-answering, SQL generation needed for data retrieval, and tabular analysis. He’s very active in the ML community, holds several patents, and has published in top ML conferences, including KDD.

Chloé Messdaghi: Closing Remarks (5 minutes) - 11:30am PT | 2:30pm ET | 6:30pm UTC/GMT

• Chloé Messdaghi closes out today’s event.