At the center of this talent exodus is Noam Shazier, who is called the father of ‘Transformer’ architecture. Shazier led Google's Gemini project and played a key role in dramatically improving the performance of AI models, and the news that he was moving to OpenAI came as a huge shock to the industry. Shazier possesses irreplaceable capabilities in drawing the blueprint for a modern large-scale language model (LLM), so much so that OpenAI's CEO Sam Altman selected him as a member of the 'dream team' that he had wanted to recruit from the beginning of its establishment. In particular, the fact that Google ultimately let him go to Open AI even though it took the initiative to acquire Character AI for $2.7 billion to bring him back suggests a serious crack in Google's AI strategy. Shazier will now oversee model architecture research at OpenAI and focus on designing next-generation pre-training technologies for deeper and more stable intelligence expansion.

Google's losses don't stop there. John Jumper, winner of the 2024 Nobel Prize in Chemistry, also shook up the industry by leaving Google DeepMind and joining Anthropic. His change of job, which changed the paradigm of life science and new drug development through AlphaFold, revealed Anthropic's will to occupy a unique position in the field of ‘general artificial intelligence (AGI)’ that goes beyond simple text generation models and solves scientific challenges. Although Google has a huge business model of search engine and advertising revenue, it is pointed out that it is losing out to companies such as OpenAI and Anthropic that present a flexible organizational culture that can focus only on AI research and a clear vision of implementing superintelligence. The phenomenon of key researchers moving to competitors one after another is not simply a matter of treatment, but also proves that differences in research autonomy and technological vision are what determine the choices of talented people.

From a technical perspective, the recruitment of Noam Shazier is a key puzzle to complete OpenAI’s ‘intelligence expansion’ strategy. Instability of logit values ​​or memory bottlenecks that occur in the process of training a large model are high-level engineering problems that cannot be solved by simply increasing parameters. Shazier has demonstrated outstanding ability to maximize the expressiveness of models and utilize hardware resources to the extreme through cutting-edge architectural technologies such as SwiGLU activation function and RMSNorm. Through this, OpenAI is pursuing ‘pre-training innovation’ that goes beyond simply inputting a lot of data, stabilizing the mathematical terrain within the model and dramatically increasing learning efficiency. This also has the effect of sending a strong signal to investors that OpenAI is further solidifying its technological moat ahead of the upcoming IPO.

Along with the movement of talent, each company's policy actions are also becoming busier. In addition to recruiting technical talent like Shazier, OpenAI is also focusing on managing policy risks and responding to long-term social influence, such as recruiting Dean Ball, who was an AI advisor to the Trump administration, to establish a ‘strategic future’ team. This is a strategy to proactively manage the impact that superintelligence AI will have on the labor market and government relations. Meanwhile, Google is trying to maintain its technological pride by continuously releasing new products such as Gemini 3.5 Flash despite the painful reality of talent loss. However, attention around the world is focused on whether Google, which lost its core brains, will be able to regain its lost AI leadership, or whether a new order led by Open AI and Anthropic will reorganize the market.