25 years at Google: from PageRank to AGI

Google Execs: Jeff Dean & Noam Shazeer

Credit and Thanks: 
Based on insights from Dwarkesh Patel.

Key Learnings

  • Embrace integration by developing products that combine multiple functionalities for a seamless user experience.

  • Invest in specialized hardware to enhance computational efficiency and performance in AI applications.

  • Foster a culture of continuous learning within your team to drive innovation and adaptability.

  • Prioritize ethical considerations in technology development to align with societal values and mitigate risks.

  • Anticipate future compute demands and design scalable infrastructure to accommodate growth.

Today’s Podcast Host: Dwarkesh Patel

Title

25 years at Google: from PageRank to AGI

Guests

Jeff Dean & Noam Shazeer

Guest Credentials

Jeff Dean is a renowned computer scientist and software engineer, currently serving as Google's Chief Scientist and co-leader of the Gemini AI project at Google DeepMind. His career at Google, which began in 1999, includes pivotal contributions to foundational technologies like MapReduce, Bigtable, TensorFlow, and Spanner. Dean holds a Ph.D. in Computer Science from the University of Washington and has been recognized with numerous accolades, including election to the National Academy of Engineering.

Noam Shazeer is an influential computer scientist and entrepreneur best known for co-authoring the seminal paper "Attention Is All You Need," which introduced the transformer model that underpins modern AI systems. He co-founded Character.AI in 2021 after leaving Google, where he had developed key technologies like Google's spelling corrector and the Meena chatbot. In 2024, Shazeer returned to Google to co-lead the Gemini AI project as part of a $2.7 billion deal involving Character.AI's technology.

Podcast Duration

2:15:35

Read Time

Approx. 5 mins

Deep Dive

Jeff Dean and Noam Shazeer reflected on their extensive journeys at Google, which began for Dean in 1999. He recounted the early days of the company, a time when it was still a small startup with a handful of employees. This intimate environment fostered a culture of collaboration and innovation, where every team member was deeply involved in the company’s mission. For startup founders, this highlights the importance of building a strong, cohesive team in the early stages of a venture. Founders should prioritize creating an inclusive culture that encourages open communication and collaboration, as this can lead to innovative solutions and a shared vision.

Dean addressed the future of Moore's Law, noting that while traditional CPU improvements have slowed, the rise of specialized hardware like Tensor Processing Units (TPUs) is paving the way for significant advancements in computational efficiency. This shift is crucial for founders to understand, as it suggests that investing in specialized technology can yield better performance for AI applications. Founders should consider how they can leverage specialized hardware to enhance their products and services, ensuring they remain competitive in a rapidly evolving market.

Dean’s undergraduate thesis on parallel backpropagation serves as a testament to the value of foundational knowledge in driving innovation. He explored the intricacies of neural networks and their training processes, which laid the groundwork for his later contributions to AI. This experience underscores the importance of deep technical expertise for founders and their teams. By fostering a culture of continuous learning and encouraging team members to deepen their understanding of core technologies, founders can drive innovation and stay ahead of the curve.

The discussion then shifted to the emergence of large language models (LLMs) in 2007, a pivotal moment in AI development. Dean and Shazeer shared their “holy shit” moments when they realized the transformative potential of these models to fulfill Google’s original mission of organizing the world’s information. For founders, this serves as a reminder to remain open to unexpected breakthroughs and to recognize the potential of their innovations. Founders should cultivate an environment that encourages experimentation and exploration, allowing their teams to pursue bold ideas that could lead to significant advancements.

Dean elaborated on the concept of doing search in-context, which enhances the user experience by providing more relevant and personalized results. This approach is indicative of the broader trend towards integrating AI into everyday applications, a strategy that founders should adopt. By focusing on user-centric design and leveraging AI to improve their products, founders can create more engaging and effective solutions that resonate with their target audience.

Looking ahead, Dean speculated on what models in 2027 might achieve, suggesting that we could see a new architecture every day, driven by automated chip design and an intelligence explosion. This rapid evolution necessitates that founders remain agile and adaptable, ready to pivot their strategies as new technologies emerge. Founders should invest in research and development to stay informed about the latest advancements and be prepared to integrate them into their offerings.

The conversation also touched on the future of inference scaling, with Dean noting that Google is already conducting multi-datacenter runs to enhance computational efficiency. This insight is particularly relevant for founders, as it highlights the importance of scalability in their operations. By designing systems that can efficiently handle increased demand, founders can position their startups for long-term success.

Debugging at scale emerged as another critical theme, with Dean emphasizing the challenges of maintaining performance and reliability in large systems. For founders, this underscores the necessity of implementing robust testing and debugging processes from the outset. By prioritizing quality assurance and investing in tools that facilitate effective debugging, founders can mitigate risks and ensure their products meet high standards.

Reflecting on the fun times at Google, Dean and Shazeer highlighted the camaraderie and collaborative spirit that characterized their early experiences. This sentiment is a valuable lesson for founders, who should strive to create a positive and enjoyable work environment that fosters creativity and collaboration. By investing in team-building activities and promoting a culture of support, founders can enhance employee satisfaction and drive innovation.

As they discussed the world compute demand in 2030, Dean emphasized the need for scalable solutions to meet the growing demand for AI capabilities. Founders should anticipate this trend and plan for future growth by investing in scalable infrastructure and exploring modular approaches that allow for flexibility and adaptability.

The conversation also touched on the importance of keeping a giga-MoE (Mixture of Experts) in-memory, which allows for efficient processing of large models. This insight highlights the need for founders to consider the architecture of their systems carefully, ensuring they can handle the demands of modern AI applications. By prioritizing efficient design and leveraging advanced architectures, founders can enhance the performance of their products.

The discussion on open research highlighted the pros and cons of sharing knowledge within the AI community. Dean and Shazeer acknowledged that while open research fosters collaboration and accelerates innovation, it also raises concerns about competitive advantage and the potential misuse of technology. For founders, this underscores the importance of striking a balance between transparency and protecting intellectual property. They should consider how to engage with the research community while safeguarding their unique innovations. This could involve participating in collaborative projects or sharing findings that contribute to the broader field without compromising their competitive edge.

As the conversation progressed, the theme of “going the distance” emerged, emphasizing the need for sustained effort and commitment in the pursuit of AI advancements. Dean reflected on the long-term vision required to achieve significant breakthroughs, noting that progress often comes from years of incremental improvements rather than overnight successes. For startup founders, this serves as a crucial lesson in perseverance. They should cultivate a mindset that values long-term goals and incremental progress, understanding that the path to innovation is often winding and requires resilience. By setting realistic milestones and celebrating small victories along the way, founders can maintain motivation and drive their ventures forward.

Actionable Insights

  • Build a strong foundational knowledge base within your team to drive innovation.

  • Integrate AI technologies into your products to streamline operations and improve user experiences.

  • Create a culture that prioritizes ethical considerations in technology development.

  • Develop scalable infrastructure to accommodate future growth in AI demand.

  • Explore new techniques for knowledge distillation to improve AI efficiency.

  • Consider the implications of open research on your competitive strategy.

Key Quote

“AI fulfills Google’s original mission of organizing the world’s information, and as we look to the future, the potential for these technologies to transform industries is immense.”

As we move towards 2030, the demand for AI capabilities will surge, necessitating advancements in specialized hardware and scalable solutions. The rapid evolution of AI architectures will likely lead to daily innovations, pushing startups to remain agile and responsive to new developments. Ethical considerations will become increasingly critical as AI systems grow in power, prompting founders to implement robust safeguards against misuse. The landscape of AI research will continue to evolve, with a focus on modularity and knowledge distillation, enabling more efficient and effective AI applications across various industries.

Check out the podcast here:

Latest in AI

1. OpenAI introduced premium AI agents with tiered pricing ($2K, $10K, $20K/month) for business professionals, developers, and researchers, powered by GPT-4.5 and backed by a $3B SoftBank investment. These agents handle tasks from basic automation to Ph.D.-level research, with projections they could generate 25% of OpenAI’s long-term revenue.

2. Google Search’s AI Mode integrates Gemini 2.0 for conversational queries, real-time data, and follow-up questions, available to Google One AI Premium subscribers. The feature balances AI summaries with web links but risks reducing traditional search traffic.

3. Alibaba’s QwQ-32B reasoning model matches DeepSeek-R1’s performance using reinforcement learning (RL) and just 32B parameters, despite competitors like DeepSeek-R1 having 671B. The open-source model excels in math, coding, and tool-based tasks while optimizing efficiency for restricted hardware environments.

Startup World

1. ManusAI, developed by Monica, claims to be the first “general AI agent,” outperforming OpenAI’s DeepResearch on the GAIA benchmark. It independently handles complex tasks like stock analysis and itinerary planning, currently in invitation-only preview.

2. EcoDataCenter raised $478M to expand sustainable data facilities in Sweden, leveraging 100% renewable energy (hydro/wind) and innovative heat recycling. The funding supports a 150MW campus in Östersund, aligning with Sweden’s 2045 carbon-neutrality goals.

What did you think of today's email?

Your feedback helps me create better emails for you!

Loved it

It was ok

Terrible

Thanks for reading, have a lovely day!

Jiten-One Cerebral

All summaries are based on publicly available content from podcasts. One Cerebral provides complementary insights and encourages readers to support the original creators by engaging directly with their work; by listening, liking, commenting or subscribing.

Reply

or to participate.