Earlier this week, a panel of experts from various sectors of the AI supply chain gathered at the prestigious Milken Global Conference in Beverly Hills for a riveting discussion. The panel included Christophe Fouquet, the CEO of ASML, a company essential for creating modern chips; Francis deSouza, the COO of Google Cloud overseeing a massive infrastructure investment; Qasar Younis, the CEO of Applied Intuition, a leading physical AI company; Dimitry Shevelenko, the chief business officer of Perplexity, an AI-native search-to-agents company; and Eve Bodnia, a quantum physicist challenging the traditional AI architecture at her startup, Logical Intelligence.
The conversation covered a range of pressing topics in the AI industry, starting with the bottlenecks faced by companies due to chip shortages. Fouquet highlighted the challenges in chip manufacturing, predicting a supply limitation for the next few years. DeSouza emphasized the growing demand for chips, with Google Cloud’s revenue surpassing $20 billion in the last quarter. Younis shed light on the constraint of gathering real-world data for AI training, emphasizing the importance of physical observation over synthetic simulation.
Energy emerged as another critical issue, with DeSouza mentioning Google’s exploration of space-based data centers to address energy constraints. He explained the complexities of cooling systems in space and the efficiency gained through integrated AI stack engineering. Fouquet echoed the sentiment, noting the price of increased compute power in terms of energy consumption.
Bodnia offered a unique perspective by focusing on energy-based AI models, which aim to understand underlying data rules rather than predicting sequences. She emphasized the adaptability and efficiency of her models compared to traditional large language models. This approach is particularly suited for domains like chip design and robotics that require a deeper understanding of physical rules.
Shevelenko discussed Perplexity’s transformation into a “digital worker” concept, where their latest offering, Perplexity Computer, acts as a team member directed by knowledge workers. He highlighted the potential of this model while addressing concerns about control through the concept of granularity.
Overall, the panel’s insights shed light on the challenges and innovations driving the AI industry forward, from chip shortages to energy efficiency and novel approaches to intelligence and control in AI systems. The conversation at the Milken Global Conference offered a glimpse into the future of AI technology and the evolving landscape of the AI supply chain. Enterprise administrators have the ability to specify the level of access that agents have to connectors and tools. This includes determining whether permissions are read-only or read-write, which is crucial when agents are operating within corporate systems. When Comet, the computer-use agent developed by Perplexity, takes actions on behalf of a user, it first presents a plan and seeks approval. While some users may find this process cumbersome, Shevelenko emphasized its importance, especially after experiencing the conservative approach of a Chief Information Security Officer at Lazard, a company built on client trust over 180 years. He stressed the significance of granularity in maintaining good security practices.
Moving on to a more geopolitical perspective, Younis brought up the entanglement of physical AI and national sovereignty, a concept that was not as prevalent with purely digital AI. The deployment of autonomous vehicles, defense drones, and other physical AI technologies raises concerns about safety, data collection, and control within a nation’s borders. Younis noted that many countries are wary of having intelligence in physical form controlled by foreign entities, highlighting the complexity of these issues.
Fouquet added to this discussion by pointing out the limitations faced by Chinese AI progress due to constraints in manufacturing advanced semiconductors. Despite advancements in software, the lack of access to cutting-edge hardware puts Chinese chipmakers at a disadvantage. This contrasts with the United States, which has the necessary resources and talent to excel in the AI field.
As the panel delved into the impact of these technological advancements on the next generation’s critical thinking skills, the speakers expressed optimism. DeSouza highlighted the potential for solving complex problems with more powerful tools, such as addressing neurological diseases and advancing sustainable energy solutions. Shevelenko emphasized the accessibility for launching independent projects with tools like Perplexity Computer, noting that curiosity and agency are the only limitations.
Younis raised the issue of labor shortages in industries like farming and mining, where physical AI is filling a void rather than displacing willing workers. He highlighted the shift towards automation in these sectors and the need for continued innovation to address labor challenges.
Overall, the panel discussion shed light on the intricate intersection of technology, security, and geopolitics, emphasizing the need for careful consideration and proactive measures in navigating the evolving landscape of AI and its implications on society.