The 2022 TransformX conference may now be behind us, but you can still catch all of our sessions on-demand! Nearly 32,000 virtual and in-person attendees joined the 3-day event, which included almost 80 sessions and 108 speakers.

From navigating an AI-enabled future to putting people on Mars with ML, Alexandr Wang and the Scale team were joined by the world's top industry leaders, researchers, and practitioners of AI and Machine Learning.

Here are just a few highlights from several of the many informative sessions over the course of the 3 days.

Day 1 Highlights

The Future of Foundation, Large Language, and Generative Models

Greg Brockman, CEO, and Founder of OpenAI, joined Alexandr Wang, for a fireside chat discussing the impact of foundation models like GPT-3 and DALL·E 2 in research and enterprise.

Brockman says that foundation models can enable businesses to benefit from the cutting-edge research of institutions like OpenAI. We don’t yet know the breadth of applications that businesses will build on top of Large Language Models and Foundation Models, but he says every business will build on large, canonical, generalized models. Looking forward, employees will likely partner with machines to produce code, imagery, and text that are more materially interesting than they might have done without assistance.

In the future, Brockman believes that AI tech will be everywhere, and there will be lots of value delivered with open source models integrated into every business that people are building all types of applications on top of. He emphasized the need to bring everyone along for this AI transition but is cautious about misuse at a societal and technical level.

The State of AI Adoption

Every vertical in the world is becoming software-driven and AI is at the forefront of this incredible growth, said Thomas Kurian, CEO of Google Cloud. In his fireside chat with Alexandr Wang, Kurian emphasized productivity being the long-term primary driver for this growth with AI leading the next evolution of this.

Kurian sees incredible opportunities in applications such as AI algorithms for financial fraud, providing clean data sets for factory floors in manufacturing, and developing learned behaviors for eCommerce in a global model. He feels passionate about pushing technology innovation really far while bringing it easily to everyone on a completely open platform.

Keynote: Applying AI to Redefine Every Industry

In Alexandr Wang’s keynote, the CEO and Co-Founder of Scale discussed how foundation models are the most exciting advancement happening in AI today. With new advancements every day, these models are giving companies the ability to deliver sustainable value over time.

Wang says we’ve seen the potential for AI to sift through a wealth of unstructured data and drive astronomical ROI, by enabling innovative and novel strategies. As a business, once you’ve established the ROI, AI enables paradigm shifts that were previously unimaginable.

In line with Scale’s mission to accelerate the development of AI applications, Wang announced exciting updates to the Scale platform including:

Scale NLP - added support to power NLP models within Scale Rapid
Scale Studio - providing organizations with the capability to run high-quality, efficient labeling projects with their own team of experts
Prodigy (coming soon) - a platform of foundation models built for strategic competition and enabling asymmetric advantages

Navigating an AI-Enabled Future

Gerrit De Vynck, AI and algorithms reporter for The Washington Post, moderated a fireside chat between Eric Schmidt, former CEO and Chairman of Google and Co-founder of Schmidt futures, and Alexandr Wang.

In regards to navigating an AI-enabled future, Wang recognized the speed at which AI development is moving. Wang touched on the fact that we have to reinvent ourselves in this new paradigm shift where AI is now an underlying platform technology, rather than an application technology.

Schmidt believes that diffusion models are particularly interesting since they’re launching a broadly impactful paradigm shift. He’s excited about the capability to analyze an art form or body of information, learn from it, and then have the ability to generate it.

The importance of AI ethics being consistent with democratic principles is another point Schmidt touched on. More broadly, he said there is an ongoing race to set global standards for privacy, data rights, and censorship—it’s important to align these new standards with democracy and western values.

Redefining and Empowering Humanity in the Age of AI

In this engaging fireside chat between Lila Tretikov, Deputy CTO of Microsoft, and John Maeda, CTO of Everbridge, the two executives discussed how AI impacts humanity and how to best leverage this technology to grow.

Tretikov said creative arts has a heavy influence in large language models (LLM) and liberal arts majors will soon be able to contribute to computation in a special, meaningful way. She said AI is starting to communicate in not just a logical state but in an emotional state, and thus specific forms of artistic expression will become more accessible.

They also discussed how AI mirrors humanity and the importance of being intentional about what data we use to train our models to enhance ourselves best. We think of AI as learning from us, but in its growing capacity to extend beyond any person's memory, computing, or even historical lifetime, it enables us to explore the sum of all human knowledge.

A Vision for Advancing the Democratization of AI

We wrapped up day 1 of TransformX with an enlightening fireside chat between Alexandr Wang and Emad Mosatque, CEO of Stability AI, on emerging trends for open source AI infrastructure, the importance of data for real-world applications of AI, and predictions on the development of “text-to-everything” in artificial intelligence.

Mostaque said that Stability AI is working to enable human potential by developing Large Language Models and foundation models that are readily accessible to the broadest audience of engineers and enthusiasts. He feels that every country should have its own dataset and models and highlighted the importance of every single group, whether cultural, geographical, or activity-centric, should be represented within those data sets and models. Cultural nuances shouldn’t divide us, said Mosatque but instead will augment our potential. In order for AI to serve the common good, everyone on the planet should feel that they have access to a representative dataset and model.

Scale and Stability AI announced a partnership in which we are aiming to democratize as much of this as possible by combining Scale’s data engine and all the foundation models being built on Stability AI to enable human potential.

Day 2 Highlights

Meta’s Journey to AI-Centric: From PyTorch to Data Quality and Beyond

Mike Schroepfer, Senior Fellow and Former CTO at Meta, sat down with Alexandr Wang to discuss his time leading groundbreaking work in computer vision at Meta, natural language processing, the metaverse, and lastly his focus today on combating climate change with technology.

Schroepfer, also known as “Schrep,” says the factors that are influencing advances in CV and VR technologies include:

Ability to harness increasing amounts of compute
Ingesting much larger datasets
Algorithmic or network innovation

Schrep is proud to “push wherever possible” on AI advances that move the state of the art forward for environmental preservation in the face of climate change as well as increasingly useful mixed reality wearables. He thinks advancements in technology have the best opportunity to take people out of poverty and advance the human condition at large across the world.

Paving the Path to Generalized Robotics

A Professor at UC Berkeley and the Director of the Berkeley Robot Learning Lab, Pieter Abbeel describes a four-step framework that might lead, ultimately, to generalized robotics.

He believes this is possible through a combination of different data sources:

(99%) The biggest data source will be internet data
(0.99%) Simulator (mostly) and real-world (little) data
Fine-tuning with human-in-the-loop

Abbeel believes that if we build the requisite systems to support generalized robotic applications, we’ll end up developing a single large neural network that is ready to learn new things quickly, including applications outside robotics.

How DeepMind Is Paving the Way From AI Research to Real-World Impact

Koray Kavukcuoglu, Vice President of Research at DeepMind, sat down with Alexandr Wang to discuss how the DeepMind researchers consistently bridge industry, science, and AI domains to create the world’s most meaningful AI advancements.

Kavukcuoglu shares his belief that the biggest applications of AI must be the things that benefit the world the most. Accordingly, we need to think about an environment where these models are value-aligned with either large populations or humanity as a whole. Kavukcuoglu feels it’s critical that people with diverse backgrounds contribute to the development of these models which yields the most useful technology.

He’s excited about a future where AI:

Is accomplishing all that we expect from it right now
Is a core technology that is part of everyone’s daily lives but is also making a positive impact
Becomes a huge positive differentiator in everything that we do

Lessons From Scaling Multiple Billion-User Apps at Google

In the engaging fireside chat with Bradley Horowitz, VP and Advisor at Google, they discussed the potential of AI-enabled features and applications being explored further and the need to solve the simple problems before being able to convince customers to try more sophisticated features.

Horowitz is excited about the future of AI impacting the shipping, logistics and healthcare industries and sees immediate product opportunity for DALL·E 2. He stressed the importance of bringing in AI into the product development life cycle at the right moment in order to provide the right value-add—that it should empower the product and not be the product.

Inside Amazon’s AI-Powered eCommerce Growth With Jeff Wilke

“AI is fundamental to doing anything at scale effectively,” said Jeff Wilke, former Amazon executive and Co-founder & Chairman of Re-Build Manufacturing. In his fireside chat with Alexandr Wang, he discussed the biggest trend being the consolidation of these various learning models into a system.

Wilke shared that in eCommerce, AI will play a huge role in marketing, discovery, transactional experience, shipping logistics and more opportunities to do topology strategy work. He believes that a fundamental leap in the efficiency of today’s supply chains will free up significant capital for further investment.

Building Inclusive AI for 400 Million Users at Pinterest

Pinterest is committed to content diversity and to building an inclusive experience on the platform. Nadia Fawaz, Senior Staff Applied Research Scientist and Technical Lead of Inclusive AI at Pinterest says the top request from Pinners is that they want to feel represented in the product.

Developing inclusive AI in the product requires a multi-disciplinary approach, not just technical skills.

An inclusive AI model involves:

Collecting diverse data
Training the AI model with bias mitigation
Serving diverse and relevant results

It’s more important than ever before to design inclusive systems that remove historical biases and to change technology with intent.

Day 2 Federal Track Highlights

Why the U.S. Must Win the Global Innovation Race

In Admiral Bill McRaven’s discussion with Alexandr Wang, he discusses how the U.S. can maintain its edge in innovation and international policy.

He emphasizes the importance of determining how we need to be faster in making decisions to enable the utmost technological capability. Making critical decisions in this domain requires connecting the technology dots and continuously assessing targets.

Admiral McRaven touches on how the military will always need the platforms and the budget will be weighted towards that, but we’ll begin to spend more and more on software, sensors and technology that enable the platforms.

The Future of AI in America: A View from The White House

Dr. Lynne Parker, Associate Vice Chancellor at the University of Tennessee, Knoxville (UTK), and Director of AI Tennessee Initiative, joined Scale's Managing Director and Head of Strategy, Michael Kratsios for an in-depth fireside chat on the future of AI. Dr. Parker touched on the importance of needing to bring together not only the people who know AI, but especially the experts close to each use case and domain, particularly those with geographic diversity.

In terms of the future of global AI policy, Dr. Parker emphasizes the importance of the development of an AI risk management framework (methodology). These collaborative efforts such as the AI center of excellence or the AI community of practice brings these agencies together to share ideas, best practices and helps accelerate the adoption of AI across the federal government. But she believes that without the talent and information exchange it will be slower.

Day 3 Highlights

Extending AI's Benefits to Society as a Whole

In a riveting conversation between James Manyika, SVP for Technology and Society at Google, and Alexandr Wang, James discusses how prioritizing utility and focusing on harm reduction should be at the core of principled use and deployment of AI. He shares not only how to make these systems better but how we can effectively implement these systems out in the real world. He feels these are often questions related to social technical embedding and how to do that effectively, whether it’s the evolution of our institutional systems or the evolution of how society itself works to adapt and take advantage of these capabilities.

Learn more about his thoughts on how we need to adapt to the new forms of work and activities that evolve as these systems become more sophisticated.

What’s Next for ML? Trends from Powering Alphabet’s Production ML

Francois Chollet is a driving force behind the Keras deep-learning library and TensorFlow, facilitating the rise of machine learning. In his keynote, the Software Engineer and AI Researcher at Google shares what he considers to be the most important machine learning trends and their broader implications.

These trends include:

Democratization & industrialization
Larger models, trained on more data
ML moving into the real world
Increased focus on privacy and safety

Chollet believes that over time ML will make its way into every problem where it can help. He believes that it will become as commonplace as web development today.

Scalable AI Solutions for Driverless Vehicles

Motional’s CTO, Laura Major, discusses her company’s ML-first approach to building the entire AV stack. She talks through best-in-class techniques to use the right data and infrastructure to fuel innovation in driverless vehicles. Through these techniques, Major believes she and her team will build truly autonomous vehicles faster than the rest of the industry.

Major emphasizes the importance of scale over domain expertise, especially in the autonomous vehicle industry. With the release of NuScenes, the world’s first publicly available real autonomous driving dataset, Motional has pioneered a data sharing culture that has now extended across the industry.

The IONIQ 5 Robotaxi, Motional’s all-electric, next gen vehicle, is the launch of the first fully driverless ride-hail service in LV in 2023.

AI and the Future of NFT Marketplaces With OpenSea

The rise of NFTs has exploded over the past year and the NFT market size is expected to grow by over 140 billion over the next 5 years. During this fireside chat Shiva Rajaraman, VP Of Product at OpenSea, was joined by Vijay Karunamurthy, Head of Engineering at Scale, where they discussed building trust and safety on consumer platforms, using operational AI to scale new markets, and the future of NFT marketplaces.

OpenSea is focused on security features and protecting users from issues that can crop up with NFTs. To aid in trust and safety, Rajaraman said one of the key things OpenSea is doing with ML, is making sure they can detect copycats at scale. It’s often difficult to delineate between a copy of something versus a valid remix and OpenSea is working on training models that can detect this.

Rajaraman is excited about a future where everyone can finally create and contribute to creative platforms that become their community in many ways. And because we have many more creative ideas and ways to get into that system, we’ll see these things flourish and eventually end up turning into the future great brands of the world.

At a practical level, Rajaraman would love to see that everyone is part of a community that values their creative expression and contributions to it. Where there’s a place you can go to contribute to the world’s work and see that work rewarded back to the community.

Building Amazon Astro: The First Multi-Purpose Home Robot

Last year, Amazon announced Astro, a new household robot for home monitoring that brought together advancements in AI, computer vision, sensor technology, and voice and edge computing. Astro brought new advances to the consumer robotic market around human-robot interaction and multimodal AI, just as the online shopping market became supercharged by the pandemic’s shelter-in-place orders.

In this fireside chat, with Vijay Karunamurthy, Head of Engineering at Scale, Dr. Ken Washington, VP of Software Engineering for Consumer Robotics at Amazon, discussed scaling robotics at Amazon and new features in Astro that are novel in the areas of computer vision, perception training, and mapping.

Dr. Washington feels passionate about building features in Astro that are more designed to support aging in place, elder care and people with disabilities. Interacting with pets and kids and having more utility around the home round out the big three features he is excited about.

Check out these sessions and more by visiting AI Exchange.