‘Stargate’ venture to invest $500 billion in AI infrastructure // new re-industrialization
Millions of data workers are implicitly or explicitly involved in AI training // eventually all humans will do
Three leading technology CEOs — Larry Ellison of Oracle, Masayoshi Son of SoftBank, and Sam Altman of OpenAI — announced the formation of a new company called Stargate. This initiative will invest $500 billion in AI infrastructure in the United States, creating over 100,000 jobs and building massive data centers and energy generation facilities. The project is described as the largest infrastructure investment in history and aims to position the U.S. as a global leader in AI and technology. The CEOs emphasized the project’s potential to revolutionize healthcare, including cancer treatment and disease prevention, while also addressing national competitiveness, particularly against China. The announcement was made with strong support from the U.S. government, highlighting the project’s alignment with national priorities and its potential to drive economic and technological advancements.
President Trump announced a new joint venture called Stargate, aimed at building and growing a structure in the U.S., with Oracle committing $100 billion to start, potentially reaching up to $500 billion over four years. Microsoft is one of the partners in this venture. During an interview, Microsoft addressed its partnership with OpenAI, emphasizing that despite OpenAI expanding to other platforms beyond Microsoft’s Azure, the partnership remains strong. Microsoft highlighted its exclusive access to OpenAI’s models and its commitment to supporting OpenAI’s growth, including significant Azure consumption.
The discussion also touched on competitors like Salesforce CEO Marc Benioff, who criticized the partnership, suggesting a potential long-term disconnect between Microsoft and OpenAI. Microsoft countered this by reaffirming its commitment to OpenAI and its role as a technology partner. Additionally, Microsoft clarified its investments in AI, noting it spends $80 billion annually on capital, with a focus on building Azure infrastructure and supporting OpenAI’s models globally.
The conversation also explored the challenges of scaling AI models, with Microsoft acknowledging that while scaling laws are still relevant, advancements in computing power and algorithms are driving progress. Microsoft is also developing its own AI models, such as small language models (SLMs), but remains focused on leveraging its partnership with OpenAI to deliver value to customers. The interview concluded with a discussion on the importance of processing power and algorithm breakthroughs in advancing AI capabilities.
PATH TO AGI
In this episode of the Big Technology Podcast, Demis Hassabis, CEO of Google DeepMind, discusses the journey towards Artificial General Intelligence (AGI) and the current state of AI research. Key points include:
Progress Towards AGI: Hassabis notes significant progress in AI over the past decade, but believes AGI — systems capable of human-like cognitive abilities — is still a few years away. Current AI models, while advanced, lack capabilities like reasoning, long-term memory, and hierarchical planning.
Current AI Limitations: Today’s AI models excel in specific tasks but are inconsistent and flawed in others. Hassabis emphasizes the need for AI to exhibit robust, consistent behavior across all cognitive tasks to achieve AGI.
Creativity and Invention: A true AGI should be able to invent new scientific hypotheses, not just prove existing ones. Current systems are far from this level of creativity.
AI Products and Future Developments: AI products like Gemini are useful for niche tasks but are not yet pervasive in daily life. Future developments aim to create universal assistants that are more integrated into everyday activities.
Mathematics and Reasoning: AI systems are making strides in mathematics and reasoning but still make basic errors. A truly general AI would be consistently strong across all domains.
Building World Models: AI needs to develop accurate models of the world to understand and interact with it effectively. This is crucial for applications like robotics and universal assistants.
Deception in AI: Hassabis expresses concern about AI systems developing deceptive behaviors, which could undermine safety tests and evaluations. Secure sandboxes are proposed as a way to test and prevent such behaviors.
Future of the Web: As AI becomes more agent-based, the web will evolve to facilitate interactions between AI agents, potentially changing how we interact with websites and apps.
AI in Science: DeepMind’s work on decoding protein folding has won a Nobel Prize, and the next goal is to create a virtual cell model to simulate biological processes, accelerating drug discovery and other scientific research.
Superintelligence: Hassabis envisions a future where AGI and superintelligence coexist with humanity, potentially leading to a flourishing society as depicted in science fiction. He stresses the need for philosophical guidance to navigate this future.
Overall, Hassabis highlights both the exciting potential and the significant challenges in the journey towards AGI and beyond.
