Google Shifting Focus from AI Leadership to Real-World Impact

Google's AI models are currently six months behind the state-of-the-art in coding capability, according to Semafor. The company's loss of talent, including Demis Hassabis stepping down as CEO of DeepMind and Chief Scientist Jeff Dean leaving to found a startup, may make it unlikely for Google to regain the lead in Large Language Models (LLM). However, this is not necessarily a cause for concern.

The AI landscape has shifted from focusing on who leads the LLM race to how effectively these models are used in real-world tasks. The new benchmark is simple: does it work and what did it cost? Google's Gemini diamond feature allows users to interact with its products using natural language, such as searching email and YouTube or getting AI feedback on writing in Google Docs.

The company aims to expand this concept through Gemini Spark, which ties together multiple Google products and pulls data from billions of users. In the future, these features will become more integrated with physical machines like stoves, cars, and eventually robotics.

However, using the most powerful AI models would be too expensive, even if it were possible to handle the compute power required. Sergey Brin, Google's cofounder with significant supervoting shares and status, is likely to play a key role in solving these challenges.