From: allin
Artificial intelligence (AI) is actively changing the world [00:01:38], with Sergey Brin, co-founder of Google, returning to help lead Google’s efforts in this area [00:00:20]. As a computer scientist, Brin expresses that he has never seen anything as exciting as the AI progress of the last few years [00:02:35]. He notes that AI, which was once a footnote in computer science curricula in the 1990s [00:02:47], has seen miraculous progress, especially with neural networks that were previously discarded approaches [00:03:05]. This advancement is attributed to increased compute power, more data, and clever algorithms [00:03:16]. New AI capabilities are emerging monthly [00:03:39].
Cross-Industry and Technical Field Impact
AI’s reach extends to numerous aspects of day-to-day life [00:04:12].
Software Development and Programming
AI is profoundly changing the way programming is approached [00:04:17]. Writing code from scratch now feels “really hard” compared to asking an AI to do it [00:04:28]. Brin himself has used AI to write complex code, such as an automatic Sudoku puzzle generator and solver to test AI models [00:04:56]. He notes that many engineers are not yet fully utilizing AI tools for their own coding [00:05:22]. Despite instances where AIs make “stupid mistakes,” their power to help people do things they never would have done, such as programming complicated tasks with kids, is seen as “magic” [00:16:36]. Google, initially hesitant to deploy AI models like those for code writing due to potential errors, has since become more willing to take risks and embrace “embarrassments” to release this powerful technology [00:15:13], [00:17:18].
Search and Information Retrieval
Conversational AI tools and Large Language Models (LLMs) are seen by some analysts as a potential threat to Google Search [00:01:53]. However, Brin suggests that AI is an extension of search and a way to retrieve information [00:04:00].
Biology and Science
In biology, Google’s AlphaFold has been widely adopted by biologists and its more recent variants [00:10:54]. Additionally, a Graph Neural Net developed by Google outperformed existing physics forecasting models [00:06:42]. Google’s AI also earned a silver medal in the International Math Olympiad using a combination of formal theorem proving, geometry-specific AI, and general-purpose language models [00:07:04].
Robotics
Robotics is in a “wow stage,” with impressive demonstrations of general-purpose language models enabling robots to perform tasks with minimal fine-tuning [00:11:18]. However, achieving the level of robustness for day-to-day usefulness remains a challenge [00:11:36]. Google’s previous ventures into robotics, including Boston Dynamics, were too early as they lacked the modern AI technology that incorporates vision and image understanding [00:12:03].
Design and Creative Fields
AI models are envisioned to be capable of designing houses [00:05:41].
Evolution of AI Models
Historically, different AI techniques were used for distinct problems, such as chess-playing AI versus image generation [00:06:28]. While some advocate for “ginormous general purpose LLMs” or a “God model,” Brin notes that the trend is towards more unified or shared models and architectures, even if not a single “God model” to rule them all [00:07:48].
Compute Infrastructure and Demand
There is a massive demand for compute resources like TPUs and GPUs, leading companies to build out infrastructure as quickly as possible [00:09:33]. Google, for example, has to turn down cloud customers due to insufficient compute availability [00:09:41]. This demand is driven by both internal model training and serving, as well as external enterprise needs for running inference on AI models and developing new applications [00:09:47]. While acknowledging the rapid buildout, Brin expresses caution against blindly extrapolating compute requirements for future AI models by “three orders of magnitude” [00:10:04], partly because algorithmic improvements are potentially outpacing the need for increased raw compute [00:08:36].
Competition and Value Creation
While competition among AI companies like Google, Meta, Amazon, OpenAI, Anthropic, and Mistral is acknowledged and seen as helpful [00:18:54], Brin believes there is “tremendous value to humanity” to be created [00:19:50]. He compares the new AI capability to the advent of the internet and web, which significantly increased human capability globally [00:20:20]. He states that the “new AI is another big capability” that almost everyone in the world can access [00:20:29].