Summary of Key Points
This article discusses the changing relationship between Google and DeepMind: from a "miraculous collaboration" in 2014, where scientists led the research with Google providing funding and time (the pre-LLM era), to Google's shift towards rapid commercialization and capital returns following the rise of ChatGPT. As a result, key DeepMind scientists such as Jeff Dean and John Jumper have left the company. Although Demis Hassabis was temporarily retained, he no longer holds real power. This marks the end of the pre-LLM era, which allowed scientists to conduct research at their own pace without the pressure to generate immediate profits. The AI industry has evolved from a small circle of researchers in laboratories into a massive industrial system that requires substantial investment and focuses on short-term returns.
Detailed Analysis
1. Hassabis Didn't Leave? – The Company Retained His "Face," Not His Real Power
Hassabis intended to leave the company along with Jeff Dean, a legendary Google engineer. However, management feared that the departure of both AI leaders would harm the stock price, so they forced him to stay. They granted him the nominal titles of "Chairman and Chief Scientist of Alphabet," but his actual authority had diminished. He no longer oversees DeepMind's day-to-day operations and is only nominally in charge of AGI and drug research (Isomorphic Labs). In essence, Google kept Hassabis around to leverage his Nobel Prize reputation to reassure investors; he might still leave at some point. It's like the company keeping a "living brand" without giving him real responsibilities, as the actual power has been transferred to Koray, Google's chief AI architect, who understands commercialization better.
2. The "Miraculous Collaboration" of the Pre-LLM Era: Google Funded, DeepMind Focused on Research
When Google acquired DeepMind in 2014, they agreed to a flexible arrangement:
- DeepMind would retain its London laboratory and independent research culture, without the pressure to make immediate profits.
- Google promised to establish an AI ethics committee to oversee technical safety.
- Google would provide funding and computing resources for years-long projects (such as AlphaFold, which took several years to solve protein structure problems).
At that time, Google was a large company with the financial means to afford such investments. They were willing to share AlphaFold freely with 3 million scientists worldwide without any commercial gain, as they were interested in the potential of AGI rather than short-term profits.
3. The Arrival of LLMs and the Need for Profit
With the success of ChatGPT, everything changed. AI is no longer just a theoretical endeavor in laboratories; it has become a profitable product:
- Google is spending $200 billion annually on computing power and data centers.
- Buffett invested $10 billion in Alphabet, asking whether this investment would generate more money.
- Sundar Pichai repeatedly emphasized the need to "accelerate" and "act quickly" to develop the Gemini model.
Now, Google's focus is on generating revenue from AI: buying chips, building data centers, and pushing scientists to produce market-ready products. It's like a boss giving you $1 million and demanding a profitable product within three months—no more room for slow, experimental approaches.
4. Scientists Leaving the Old Era: They Seek Freedom, Not KPIs
The key scientists from the pre-LLM era can't handle the current commercial pressure and are leaving:
- Jeff Dean (co-founder of Google Brain) started a nonprofit to conduct research that may not align with company interests.
- John Jumper (creator of AlphaFold) joined Anthropic, a competitor of OpenAI.
- Noam Shazeer, author of the Transformer paper, moved to OpenAI.
Their reasons are simple: in the past, they could pursue research for its own sake; now, they need to generate profits. For example, AlphaFold might have been commercialized if it were developed today. These scientists want to be free from the constraints of strict KPIs and use Google's funding (such as Discovery Loop) to continue their independent research.
5. The Unreturnable 2013: AI from a Small Circle to a Massive Industry
Fourteen years ago, Geoffrey Hinton used a upside-down trash can as a computer in a hotel room to auction his company (DNNresearch). Today, AI requires millions of dollars in chips and billions of dollars in funding. Scientists have turned AI from a niche field into a massive industry:
- In the past, they could research without immediate expectations; now, they must see immediate results.
- In the past, scientists made decisions; now, capital does.
Google can delay Hassabis's departure, but it can't return to the 2013 era of relaxed research. Back then, Larry Page told Hassabis, "You don't need to create another Google." Today, Google needs AI products that generate profits, not distant AGI dreams.
In Conclusion
This article highlights the complete shift in the AI industry from an era dominated by scientists and idealism to one driven by capital and reality. The pioneers of AI are being marginalized by the industrialized system they helped create.
(End of translation)