Summary of Key Points
Google has recently made significant changes to its AI leadership structure: Demis Hassabis, co-founder of DeepMind, has shifted from CEO to Chief Scientist at Alphabet (also serving as Chairman of DeepMind), focusing on the long-term strategy and social impact of AGI (Artificial General Intelligence); Koray, the former CTO, is in charge of daily operations, responsible for the development and implementation of the Gemini model; four senior members of Google Brain, including Jeff Dean, have left to start a new company, with Google maintaining a partnership as an investor. The direct trigger for these changes was the slow progress of the Gemini model (falling behind GPT and Claude), but the underlying issue is that Google's AI scientific achievements (such as Transformer and AlphaFold) have not been effectively converted into commercial advantages, and there are inefficiencies within the organization. The aim of these adjustments is to address internal "chaos" and transition AI development from a "laboratory-driven breakthrough" to a "systematic product operation" approach, in order to regain a leading position in the AI competition.
I. Major Personnel Changes: Who Moved Up, Who Left, and Who Took Over?
The changes affect three key groups:
1. Demis Hassabis: From Manager to Strategist
He no longer oversees the day-to-day operations of DeepMind (budgeting, personnel, project progress) but is responsible for the long-term direction of AGI and the social impact of AI at the Alphabet level, as well as managing the AI pharmaceutical company Isomorphic Labs. In other words, he has shifted from leading the team to providing strategic guidance.
2. Koray: Taking Over Day-to-Day Operations with a Focus on Gemini
As the former CTO of DeepMind and Chief AI Architect at Alphabet, Koray now reports directly to Google CEO Sundar Pichai. He is responsible for the development and delivery of Gemini, as well as coordinating resources such as chips, cloud services, and consumer products. Although he is an AI scientist, his expertise lies in turning research results into usable products—specifically, making Gemini competitive with GPT and integrating it into Google's search, cloud, and Workspace services.
3. Jeff Dean and Others: Leaving to Start a New Company
Jeff Dean, one of the founders of Google Brain, along with three other core technology experts (in areas such as distributed computing, Gemini, and reinforcement learning), have left to establish Discovery Loop, an AI system designed to automate scientific research. Google did not sever ties completely but invested in them and became a cloud computing partner, essentially offloading long-term, high-risk research efforts while maintaining cooperation.
II. Why the Urgent Need for Changes? The Slow Progress of Gemini
The immediate reason is the crisis around the Gemini model:
- Google formed a special task force to work on the code for the model, but it was led by Sundar Pichai and Koray, not Hassabis, indicating dissatisfaction with the pace at which Hassabis was advancing Gemini.
- The launch of the Gemini model was originally scheduled for June but has been delayed for several months. Investors are concerned about progress and talent loss, as Google has fallen behind competitors like OpenAI and Anthropic, which continue to update their models, while Chinese companies are gaining market share with lower-cost offerings.
The deeper issue is the difficulty in transforming scientific achievements into commercial products:
Google possesses many advanced AI technologies (such as Transformer and TPU chips, AlphaFold), but its representative generative AI products (like ChatGPT) were not developed by it. The problem lies in the long process between research, training, product development, and market deployment—current large models require continuous iteration (data preparation → training → optimization → user feedback → retraining), which Google's previous organizational structure could not keep up with.
III. Impacts on Google
In the short term, Google's stock price dropped by 4% due to concerns about Gemini's progress and talent loss. The departure of Jeff Dean and others is particularly significant, as they are key members capable of developing technologies from fundamental infrastructure to cutting-edge models, representing a loss of a critical technical foundation.
In the long term, there may be changes in DeepMind's culture: In the past, DeepMind attracted scientists because it provided independent space for long-term research; now, daily management must align with Google's business goals, which might lead to research projects focusing more on commercially viable paths (such as Gemini). If this balance is not struck, Google could lose opportunities for breakthroughs that may not have immediate commercial value but could spark the next technological revolution. However, Google has a plan in place: Hassabis will continue to oversee long-term scientific efforts at the company level, and it has invested in the new company founded by the departing team to maintain research collaborations.
IV. Google's Efforts to Address Internal Chaos
The main goal of these changes is to address the "entropy increase" (increasing internal complexity and efficiency decline) within the tech giant, focusing on four key areas:
1. Separating Science from Business
Hassabis will handle long-term strategies (AGI and research), while Koray will manage day-to-day operations (model progress and product implementation), ensuring that specialists focus on their respective areas.
2. Unifying Post-Merger Teams
Although Google Brain and DeepMind merged in 2023, there was still confusion within the organization. Now, Koray is responsible for integrating research, models, applications, and tools to create a cohesive team.
3. Model-Product Cycle
Models should not be isolated APIs but should be integrated into various products (search, cloud, Android, etc.). Data and feedback from product use should be used to continuously improve the models, creating a cycle where more usage leads to better models, which in turn drives more usage.
4. Combining Internal and External Innovation
Google will focus on developing commercially viable models internally, while external partners (like Discovery Loop) will undertake high-risk research. This approach allows Google to maintain both short-term profitability and long-term breakthroughs.
V. The Impact on the AI Industry
The changes by Hassabis and Jeff Dean highlight a shift in the competition: No longer is a single scientific breakthrough enough; success depends on a comprehensive system of capabilities, including chips, data centers, training frameworks, product feedback, and security governance. The most advanced research must be transformed into stable, usable products. Google's focus indicates that the next competitive advantage will come from an "industrial system" capable of continuously producing effective models. In other words, AI competition has shifted from who can develop the best technology in the lab to who can consistently produce useful products in a commercial setting.
Conclusion
Google's reorganization aims to transform itself from a leader in AI research to a leader in AI products and systems. Only by making Gemini successful can it fully leverage its resources (chips, users, cloud services) into a competitive advantage. Whether this strategy will be successful depends on Koray's ability to drive the development of these AI technologies.