At the recent All-In Summit 2024, Google Co-Founder Sergey Brin shared his thoughts on the current state of AI and Google’s involvement in advancing this technology. Brin, who recently returned to Google to work on AI, opened by emphasizing his excitement for the “mind-blowing” progress made in the field. He noted that AI has evolved from being a footnote in computer science curriculums to becoming the center of technological innovation today, adding that he finds the monthly advancements in AI capabilities to be astonishing.
Brin explained that as a computer scientist, the progress made in neural networks and machine learning over the past decade has been particularly exciting. Brin shared how AI now touches many aspects of life, including programming itself. He revealed that he has even used AI to write code, citing an example where he had AI write a program to generate Sudoku puzzles and then solve them. Brin expressed amazement at how quickly the AI completed the task, even surprising Google’s engineers, who had not fully realized the power of the tools at their disposal.
Addressing the topic of Large Language Models (LLMs), Brin discussed the industry’s pursuit of a unified model that can handle multiple tasks, hinting at the development of what some call a “God model” or Artificial General Intelligence (AGI). He remarked that while the idea of a singular model capable of performing a wide range of tasks is alluring, the current reality still requires specialized models for certain tasks. However, Brin believes that the trend is toward a more “unified model” that could bring these capabilities together in a powerful, centralized system.
When asked about the massive buildout of AI chips and computing infrastructure, Brin confirmed that Google is expanding its compute capacity as quickly as possible to meet both internal and external demand. He acknowledged that Google Cloud customers are hungry for TPUs and GPUs, and the company is sometimes forced to turn down requests due to limited availability. Brin attributed this demand to the enterprise world’s eagerness to integrate AI into their operations, but he also cautioned against blindly extrapolating trends in compute power, stating that recent algorithmic improvements may reduce the need for exponential increases in hardware.
In discussing the applications of AI, Brin highlighted the success of AlphaFold, which has revolutionized the field of biology by predicting protein structures. He described how AlphaFold has become a staple tool for biologists around the world, underscoring AI’s transformative potential in life sciences. At the same time, Brin expressed optimism about the future of robotics while acknowledging that the technology still lacks the robustness needed for widespread day-to-day use. He shared that, in hindsight, Google’s previous ventures into robotics may have been too early, but the advancements in multimodal AI models—which integrate language, vision, and other inputs—are creating a clearer path for future applications.
Brin also spoke about the balance between risk and innovation, particularly in relation to Google’s cautious approach to deploying LLMs. He admitted that Google was initially “too timid” in rolling out its AI models despite being a leader in the development of Transformer models, which laid the foundation for modern LLMs. Brin emphasized that while these models can sometimes make “stupid mistakes”, their overall potential far outweighs these imperfections. He believes companies must be willing to accept some “embarrassments” in order to push boundaries and discover new possibilities.
Reflecting on Google’s culture, Brin acknowledged that as the company has grown, there has been a tendency toward conservatism in product development. He recounted an internal story where engineers hesitated to deploy AI-generated code because it wasn’t perfect. Brin encouraged the team to move forward with the release, arguing that AI’s magic lies in its ability to enable things that were previously impossible, and perfection should not stand in the way of progress.
Brin addressed the competition in the AI space, particularly between Google, OpenAI, and Anthropic. While he welcomed the competition, noting that it drives innovation, he also emphasized that the field is vast and full of opportunities. Brin remarked that while Google was briefly overtaken in certain AI benchmarks by competitors, the company has since regained its position at the forefront of AI innovation. He proudly shared that Google was recently ranked number one in AI language model analysis for a few weeks, underscoring the progress the company has made since the launch of ChatGPT and other competing models.
Featured Image via YouTube