GenAI Superstream: Possibilities and Pitfalls

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Thomas Nield: Introduction (5 minutes) - 8:00am PT | 11:00am ET | 4:00pm UTC/GMT

• Thomas Nield welcomes you to the GenAI Superstream.

Antje Barth–Keynote: Generative AI Trends for 2024—Dos and Don’ts (15 minutes) - 8:05am PT | 11:05am ET | 4:05pm UTC/GMT

• Antje Barth discusses trends in generative AI for 2024 and shares practical guidance on the importance of high-quality data, inference optimizations, and specialized agents. You’ll explore the nuances of model evaluation and human feedback and discover a unique perspective on the promising multimodal generative AI use cases of the future.
• Antje Barth is a principal developer advocate for generative AI at Amazon Web Services. She’s also coauthor of the O’Reilly books Generative AI on AWS and Data Science on AWS. A frequent speaker at AI and machine learning conferences and meetups around the world, she cofounded the global Generative AI on AWS Meetup and the Düsseldorf chapter of Women in Big Data. Previously, Antje worked in solutions engineering roles at MapR and Cisco, helping developers leverage big data, containers, and Kubernetes platforms in the context of AI and machine learning.

Elan Head: Certifying AI for Aviation (30 minutes) - 8:20am PT | 11:20am ET | 4:20pm UTC/GMT

• Aviation companies, from small startups to major planemakers like Airbus and Boeing, are investing heavily in the development of commercial autonomous aircraft. AI promises to be a major enabler in these efforts, helping aircraft land precisely, detect obstacles in the air and on the ground, and even interface with air traffic control. Yet aviation is a highly regulated industry, and existing software certification guidance is not suitable for assuring the safety and reliability of models developed through machine learning. Elan Head discusses how industry and regulators are tackling this problem and some of the competing philosophies involved.
• Elan Head is senior editor at The Air Current, where she specializes in covering electric aircraft, autonomy, and other emerging technologies. She has written extensively about potential applications for AI in commercial aviation and the regulatory challenges involved. Elan is an award-winning journalist whose work has been recognized by the National Press Club, among others. She’s also a commercially rated helicopter pilot and flight instructor.
• Break (5 minutes)

Emmanuel Maggiori: AI Gone Wrong—A Few Caveats Before Using Generative AI in Your Project (30 minutes) - 8:55am PT | 11:55am ET | 4:55pm UTC/GMT

• Failed AI projects commonly have one, or both, of two markers: The first is a hammer-in-search-of-nails approach in which organizations decide to use AI because it seems promising and then try to find a problem to solve with it. The second is an underestimation of AI's limitations. Emmanuel Maggiori has a lot of experience participating in failed AI projects, and he shares a few caveats to help you avoid falling into these two traps.
• Emmanuel Maggiori, PhD, is a 10-year AI industry insider. He has developed AI for a wide variety of applications, from extracting objects from satellite images to packaging holiday deals for millions of travelers every day. He is author of the book Smart Until It’s Dumb, described by The Daily Telegraph as "the most lucid layman's account of AI and its limitations."

Chris Fregly: Improving Generative AI Response Times and User Experience (30 minutes) - 9:25am PT | 12:25pm ET | 5:25pm UTC/GMT

• Chris Fregly dives into a new way of comparing and optimizing GenAI models focused on lightning-fast response times. He describes some important generative-inference metrics to monitor, such as time-to-first token (TTFT), time-per-output-token (TPOT), latency, and throughput. You’ll learn various strategies for improving these metrics using techniques like flash attention, paged attention, and continuous batching.
• Chris Fregly is a principal solutions architect for generative AI at Amazon Web Services. He holds the full complement of AWS certifications and is coauthor of Generative AI on AWS and Data Science on AWS, both for O’Reilly. A cofounder of the global Generative AI on AWS Meetup, he often speaks at AI and machine learning meetups and conferences across the world. Previously, Chris was an engineer at Databricks and Netflix, where he worked on scalable big data and machine learning products and solutions.
• Break (5 minutes)

Matthew Perez: Using ChatGPT to Build API Scripts for CAD (30 minutes) - 10:00am PT | 1:00pm ET | 6:00pm UTC/GMT

• Computer programming, even at a basic level, is a valuable tool to help automate repetitive tasks such as pulling data from one spreadsheet into another spreadsheet, formatting a document, or performing some modeling-based tasks in a CAD program. But as these needs arise, you don’t always have time to learn a new programming skill to the necessary level of proficiency. Using AI to both teach and assist in the creation of these simple routines is one solution. Matthew Perez offers some examples of how ChatGPT can help teach, assist, and create routines for computer-aided design programs Fusion 360 and SolidWorks.
• Matt Perez is a trainer and consultant who works with inventors and companies around the world to help them design products and to train employees to use various software tools. He has worked within CAD systems since the late 90s, and has also developed certifications to validate users skills. Matt is exploring AI as a learning tool to assist designers and engineers who have knowledge gaps in programming achieve the goal of creating simple scripts and macros to assist with repetitive tasks.

Natalie Pistunovich: Scaling AI Systems—Challenges and Solutions (30 minutes) - 10:30am PT | 1:30pm ET | 6:30pm UTC/GMT

• Natalie Pistunovich examines the practical challenges of scaling AI systems from proof-of-concept to production, exploring common pitfalls such as managing data quality, ensuring model performance at scale, and integrating AI with existing IT infrastructure. You'll learn strategies for overcoming these challenges, including best practices for model management, deployment techniques, and maintaining AI system performance over time.
• Natalie Pistunovich is an OpenAI developer ambassador, a Google Developer Expert for Go, and a consultant for companies that want to incorporate AI. As a professional question-asker and a passionate learner, she finds joy at the intersection of AI, code, and security. She shares that joy as a university lecturer in DevOps and a public speaker and conference and community organizer in Berlin. Natalie has worked on four different continents as an engineering manager, software and hardware engineer, developer advocate, and cofounder of a mobile startup.
• Break (5 minutes)

Gary N. Smith: AI Systems Are Still Faux Intelligence (30 minutes) - 11:05am PT | 2:05pm ET | 7:05pm UTC/GMT

• LLMs are magical in their ability to create an illusion of intelligence. The fundamental problem with LLMs and other AI systems is that they cannot distinguish between truth and untruth or between meaningful causal relationships and coincidental statistical patterns. They should only be used in situations where the costs of mistakes are small—like recommending movies. But their power has persuaded many that they can be used to make hiring decisions and medical diagnoses, approve loans, set prison sentences, and influence military strategy—situations where the costs of mistakes are high. Gary N. Smith discusses the dangers of embracing the technology too readily and relying on its judgment too heavily.
• Gary N. Smith, the Fletcher Jones Professor of Economics at Pomona College, is the author of more than 100 research papers and 17 books. His bestselling Standard Deviations: Flawed Assumptions, Tortured Data, and Other Ways to Lie with Statistics was a London Times Book of the Week. The 9 Pitfalls of Data Science (Oxford University Press, 2019), coauthored with Jay Cordes, won the 2020 PROSE award for Excellence in Popular Science & Popular Mathematics. His most recent book is Distrust: Big Data, Data-Torturing, and the Assault on Science (Oxford University Press, 2023).

Shelbee Eigenbrode: Building Effective Reasoning Agents (30 minutes) - 11:35am PT | 2:35pm ET | 7:35pm UTC/GMT

• Shelbee Eigenbrode shows you how to build an agent capable of performing complex reasoning using various techniques and models like ReAct. ReAct, short for reasoning and action, is a method that combines reasoning with the ability to execute actions using tools and APIs, allowing large language models to reason about requests and perform the necessary actions to complete the tasks. All code will be shared publicly.
• Shelbee Eigenbrode is a principal solutions architect for generative AI at Amazon Web Services, where she combines her DevOps and ML backgrounds to deliver ML workloads at scale. With a 23-year career in technology and a passion for continuous innovation, she holds six AWS certifications and more than 35 patents across various technology domains. Shelbee is also a coauthor of Generative AI on AWS and cofounder of the Denver chapter of Women in Big Data.

Thomas Nield: Closing Remarks (5 minutes) - 12:05pm PT | 3:05pm ET | 8:05pm UTC/GMT

• Thomas Nield closes out today’s event.