Summary of Key Points
Over the past week, Google DeepMind has lost four of its top AI talents: Noam Shazeer, co-author of the Transformer architecture, joined OpenAI; John Jumper, a key creator of AlphaFold, along with Jonas Adler and Alexander Pritzel, two other researchers from Gemini, have moved to Anthropic. These departures have sparked significant discussion due to their involvement in Google's core AI technologies (Transformer, large models, scientific AI, AI coding, etc.), raising concerns about the decline in Google's AI competitiveness. However, DeepMind CEO Demis Hassabis responded by stating that Google is betting on the longer-term goal of achieving Artificial General Intelligence (AGI), rather than just short-term text model competitions.
Detailed Analysis
1. The Departures of AI Industry Leaders
These individuals are not ordinary employees but pivotal figures in Google's AI efforts:
- Noam Shazeer: Co-author of the 2017 Transformer paper, which laid the foundation for all large models (such as GPT and Gemini). He returned to Google less than two years ago and quickly became a key leader in the Gemini project.
- John Jumper: Creator of AlphaFold, which can accurately predict protein structures, accelerating drug development and earning him the 2024 Nobel Prize in Chemistry. He represents Google's strength in "AI for Science."
- Jonas Adler & Alexander Pritzel: One specializes in AI coding (a highly sought-after skill), and the other focuses on model training systems; both were key contributors to Gemini.
Their departures mean that Google has lost its designers of large models, advocates for scientific AI, experts in AI coding, and specialists in model training—each of these gaps directly impacts Google's core AI capabilities.
2. Why Would Top Talents Leave?
The reason these professionals left is not due to lower pay but better opportunities:
- Prospects at Startups: OpenAI and Anthropic are not yet publicly traded, offering potentially substantial returns through pre-IPO equity. Google, being a mature company, does not offer the same level of incentive potential.
- Focus on Long-Term Goals: Both companies are focused on AGI and large models, with concentrated resources. Google's broad scope (search, cloud, hardware, AI, etc.) may lead to shifting priorities. For example, before Shazeer left, the computing resources for his project were reallocated to a London team.
- Organizational Efficiency: Startups make faster decisions and can quickly turn ideas into products. As a large company, Google's bureaucratic processes may limit researchers' freedom.
3. Public Reaction
These departures have caused a stir:
- Some criticize Gemini as mediocre and claim that Google has lost its innovation edge (with GPT-4o and Claude performing better in certain areas).
- There are concerns about Google's declining position in the AI talent race, wondering if more top talents will leave.
- Even DeepMind employees suggest that Gemini is no longer at the forefront, and Shazeer might not be the last to go.
4. Hassabis' Response
Facing these doubts, Hassabis did not back down:
- Acknowledging Talent Mobility: He acknowledged that the AI industry is highly competitive, and it's normal for top talents to move between labs. Google still retains most of its top researchers.
- Diversified Approach: He emphasized that AGI requires integrating multiple abilities (language, vision, scientific reasoning, understanding of the physical world). DeepMind has a large and diverse research team, focusing on areas where OpenAI and Anthropic are not as strong.
- Long-Term Vision: Winning in AGI depends on building intelligent systems that can understand the world, not just on short-term popularity of text models. For example, Google's work on game AI (AlphaGo) was to train algorithms for real-world problems.
5. The Big Picture: The Competition for Top Talent
These departures highlight a key aspect of the AI industry:
- The Real Resource: The competition is not just about GPUs but about people who can effectively use them. Top talents are crucial for translating resources into breakthroughs.
- Changing Priorities for Talents: They now value opportunities to pursue their interests, high returns, and efficient organizations. Startups offer these advantages.
- AGI Competition Entering a New Phase: Companies are betting on different approaches: OpenAI and Anthropic focus on text and coding, while Google focuses on multi-modal and scientific AI. The winner will be the one that best aligns with the essence of AGI.
In Conclusion
These departures do not indicate the end of Google's AI efforts but signify a reshuffle of talent in the industry. Top talents are choosing platforms that better align with their goals, and Google must prove that its long-term strategy can retain them and win the battle for AGI.