Summary of Key Points
There has been significant turmoil in Google's AI department recently: Jeff Dean, the chief scientist, has left to start his own business, while Demis Hassabis, a key figure at DeepMind, has stepped back from frontline roles (now serving as chairman and chief scientist for Alphabet), with Corrado Cavucchio becoming the new person in charge of daily management. This is not an unexpected outcome; it reflects the ongoing conflicts that arose after the merger of DeepMind and Google Brain in 2023. The merger was intended to unite efforts against OpenAI, but due to power struggles, cultural differences, and resource competition between DeepMind's headquarters in London and Google's headquarters in Mountain View, there has been a continuous loss of talent, with the balance of power gradually shifting towards Mountain View. Whether the new leadership can resolve these internal issues and help Google's AI efforts catch up with OpenAI remains a critical question.
Detailed Analysis
1. The "Ideal" vs. the "Reality" of the Merger: A Secret Battle Between Two Power Centers
The initial goal of the merger was straightforward: to combine DeepMind (London) and Google Brain (Mountain View) into one strong team, ending internal rivalries and focusing on competing with OpenAI. However, in reality, both Hassabis (from London) and Dean (from Mountain View) reported directly to Google CEO Sundar Pichai. Although they were formally part of the same organization, their teams functioned almost as two separate entities.
Over the past year, power has quietly shifted towards Mountain View: Cavucchio, who was previously the CTO in London, moved to Google's headquarters, and Borjoo, responsible for critical code development, also relocated from the UK to California. A former employee described this as the "final fallout" of the merger—power first went to Hassabis, then to Cavucchio and Borjoo, leaving Dean with diminishing influence, which ultimately led to his decision to leave and start his own business.
2. Cultural Conflict: London's Academic Atmosphere vs. Mountain View's More Aggressive Style
The working cultures of the two teams differ greatly. A former London employee criticized DeepMind for becoming more "Americanized," with a culture where louder voices dominate meetings and there is a competitive atmosphere where everyone tries to claim credit for their projects. In contrast, London had a more academic approach to collaboration. The situation was further exacerbated by a trust crisis: researchers in London could access Mountain View's project documents, but not vice versa. When projects overlapped, decisions favored the team with the more influential backers. The employee noted, "There was a strong sense of politics; many things depended on who you knew, and there was a lot of resentment between the two teams."
3. Fierce Competition for Resources: Struggling to Retain Both Computing Power and Talent
In the AI field, computing power (the necessary resource for training models) is extremely valuable, and the two teams fought fiercely over it. For example, before leaving, one of DeepMind's key researchers, Shazal, had his project's computing resources unexpectedly reallocated to the London team. Even more detrimental was the loss of talent: Google once held a significant advantage, as it invented the Transformer architecture (the foundation for generative AI) in 2017, but all eight of the original authors have since left. This year, Nobel Prize-winning researcher David Jiang (who led AlphaFold) joined Anthropic, and just two days ago, Shazal switched to OpenAI. The continuous departure of key talents indicates serious internal problems.
4. Can the New Leadership Resolve the Internal Frictions?
Cavucchio, who is now in charge of daily management, acts as a "mediator" between the two teams. He was Hassabis' close associate and had previously worked at DeepMind in London for two years before moving to Mountain View, so he understands both cultures well. Google executives have mentioned that Cavucchio has been overseeing many matters at DeepMind for some time; even Google co-founder Larry Page turns to him for assistance with implementing decisions. His strength lies in his ability to coordinate relationships within large, diverse teams with strong personalities. The outside world hopes that he can bridge the gaps between the two locations and reduce internal conflicts.
5. The Future of Google's AI: Can It Catch Up with OpenAI?
Google once had a strong advantage with the Transformer architecture, but due to internal bureaucracy and slow decision-making (especially considering the difficulty of cross-continental communication between leaders in London and Mountain View), OpenAI took the lead with ChatGPT. With these recent changes in leadership, whether the new management can streamline internal processes, retain talent, and accelerate decision-making remains to be seen. If these issues are addressed, Google might have a chance to make a comeback in the generative AI space; otherwise, it may continue to fall behind.
In Conclusion
This internal struggle is essentially a battle between "efficiency vs. innovation" and "centralization vs. decentralization." If Google wants to win in the AI race, it must first dismantle the barriers within its own organization.