AI in Action: Select Insights from MIT Media Lab Summit

Imagination in Action hosted an engaging event that brought the AI community together for a deep dive into the latest developments shaping the future of business. 'Imagination in Action: Forging The Future of Business with AI' turned into a bustling hub from morning till night, packed with AI researchers, business leaders, and innovative startups all sharing insights and sparking dialogue. A special shoutout and deep thanks to John Werner , Alex 'Sandy' Pentland , and Randall Lane of Forbes for orchestrating this dynamic gathering. Amid the flurry of activity across three floors, with four sessions running in parallel, we all found countless opportunities to connect and exchange breakthrough ideas.

The event at MIT was buzzing with sessions (running in four parallel tracks) covering a myriad of AI innovations and their impact on the future of business. I'm excited to share a slice of that here:

There were two tracks, where I had the privilege to contribute:

During my Lightning Talk titled "AI’s Third Pillar": I emphasized that while Data is the new oil (Clive Humby) and Compute is the currency of the future (Sam Altman), the critical yet often undervalued third pillar is the network. This pillar forms the backbone for the next wave of machine-generated data, with IoT devices growing from 16B+ in 2023 to almost 30B in 2027. Network also provides the very foundation of flexible connectivity that brings Data and Compute together

.

At my Panel on 'Insights from Rapidly Growing AI Companies': I joined forces with AI leaders like Shashank Dixit , CEO of Deskera, Igor J. , CEO of Pyron, and Stefanos Poulis , Chief AI R&D of Seekr. Our discussion delved deep into how AI is revolutionizing the enterprise world, scaling AI revenue, and exploring emerging opportunities. We tackled the hefty topic of in-house AI model development and its associated costs and debated the readiness of Gen AI for mission-critical applications, which sparked a lively exchange of views.

During a fireside chat, Yann LeCun , Meta 's Chief AI Scientist, shared intriguing updates and insights on the trajectory of AI development. In a timely announcement coinciding with the event, Meta unveiled Llama 3, an open source AI model, in two configurations: one with 8 billion parameters and another with a staggering 70 billion parameters. The training scale was immense, with the smaller model consuming 1.3 million GPU hours and the larger one demanding 6.4 million GPU hours, all the while digesting an extensive training diet of 15 trillion tokens.

Yann is a staunch advocate for the open-source movement, emphasizing that the continual improvement of AI infrastructure hinges on its accessibility to the broader community.

Furthermore, Yann introduced us to V-JEPA, a pioneering vision model designed to learn perceptual skills akin to a young animal. Instead of reconstructing images or relying on pixel-level predictions, V-JEPA prioritizes video feature prediction. V-JEPA, standing for Video Joint Embedding Predictive Architecture, marks a significant leap toward Yann’s ambitious goal of achieving Advanced Machine Intelligence, or what many refer to as Artificial General Intelligence.

In an interview with Vinod Khosla, a luminary in entrepreneurship and venture capital, he imparted a compelling perspective: "Robots are not robots anymore. They are learning systems."

On topic of education, Vinod's advice to the new generation was to continue to pursue computer science. I was glad to hear that my choice of college major still remains strong.

In a thought-provoking interview, Stephen Wolfram , a renowned scientist and the creator of Wolfram Alpha, weighed in on the capabilities of current AI in the realm of reasoning. He noted that while today’s Large Language Models (LLMs) are adept at anticipating what comes next in structured systems like language or code—offering 'roughly-right' predictions akin to human-style reasoning—their performance is less assured in domains that demand precision. In the meticulous world of scientific inquiry, where the margin for error narrows to a sliver, these models have yet to meet the mark. Wolfram's insights underscore the nuanced journey AI must undertake to bridge the gap from probable to precise in its reasoning prowess.

Lex Fridman , an AI researcher known for his in-depth interviews and discussions on AI, unfortunately, couldn't attend in person but contributed to the conversation with a timely video. In his message, he expressed enthusiasm about several emerging trends in AI, particularly the development of personalized Large Language Models (LLMs) with the ability to remember previous interactions. This capability represents a crucial step forward in making AI interactions more contextually rich and personally relevant.

Blue Shield of Michigan shared an AI use case that can be applied to all businesses. It has harnessed AI to streamline their contract management, unveiling a 30% overlap in IT contracts through a Contracts GPT system. This savvy integration of natural language processing has led to a remarkable $10 million in savings so far, demonstrating AI's practical value in optimizing business operations.

Spotlighting innovation from the startup sphere, we have Groq and Liquid AI making waves in hardware and software, respectively. Groq designs specialized chipset for accelerating deep learning and AI inference workloads with ultra-low latency, boasting impressive efficiency gains over traditional GPU architectures. Dinesh Maheshwari , Groq's CTO, highlighted their unique Logic Processing Units (LPU) leverage programmable assembly line architecture - compared to GPU’s hub-and-spoke architecture - which enables high token per second throughput at lower power levels.

On the software front, Liquid AI stands out as an MIT spinoff co-founded by Daniela Rus , which is pioneering the concept of liquid neural networks. These networks are aiming to be more efficient, reliable, and explainable. CEO Ramin Hasani explained that their approach is adept at handling various sequential data types, such as audio, video, and time series, potentially revolutionizing how we interact with AI across multiple domains.

I am also thankful for the lovely pre-summit reception at the Quin House, and ...

also thankful to the great post-summit reception at One?Kendall?Square:

The MIT AI summit was a powerhouse of innovation, where cutting-edge ideas met practical solutions. With a lineup that ranged from academics to executives to startups, there was no shortage of insights. I had the honor of contributing to this vibrant exchange and am already eager to revisit the sessions through the recorded content. It was an event that not only informed but also connected us, and I’m looking forward to the collaborations that will undoubtedly stem from these interactions.