OpenAI Hires AI Legend Noam Shazeer to Bolster Its Manpower

To gauge the gravity of the hiring, here’s what Sam Altman, the company’s CEO, had to say about Noam Shazeer: “Noam is one of the people I have most wanted to work with since the very beginning of OpenAI. Only took 10 years. I think it will be worth the wait!” Ahead of its IPO, OpenAI has made a couple of new hires to bolster its manpower, especially in an era of government regulations dictating the release of new AI products. Shazeer is one of them – the man who is widely regarded as an “AI Legend” and one of the core architects of modern-day AI. Shazeer announced his transition on X (formerly Twitter), bidding goodbye to his team at Google and embracing the opportunities at OpenAI . It is interesting to see Shazeer making the switch in less than two years after Google spent a jaw-dropping $2.7 billion to rehire him. At a time when the competition is cut-throat in this part of the business world, Shazeer’s switch to OpenAI is one of the most significant talent shifts we have seen, right on par with Andrej Karpathy’s move to Anthropic. But why Noam Shazeer? Why did Sam Altman decide to onboard one of Google’s top talents? Let’s find out. Shazeer’s definitive contribution to human history came in 2017, when, alongside seven other researchers at Google, he co-authored the landmark paper titled “Attention Is All You Need.” Before we dive in, you need to understand the context. Prior to this paper going online, AI models relied heavily on Long Short-Term Memory (LSTM) networks, processing language sequentially (word by word). Shazeer and his colleagues were deeply frustrated by the immense computational inefficiencies of LSTMs. Hence, they came up with the Transformer architecture – a framework that makes use of “self-attention” mechanisms to process entire sequences of data simultaneously. Shazeer’s work on the Transformer, alongside his pioneering developments in Mixture-of-Experts (MoE) architectures, enabled massive AI models to run efficiently by activating only the “experts” relevant to given tasks. Every major large language model (LLM) in existence, including OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude, is fundamentally built upon the Transformer architecture Shazeer co-invented. Back in 2000, Shazeer joined a two-year-old startup operating out of a modest office space that most people hadn’t heard of until then, Google. His major breakthrough was completely rewriting and vastly improving Google’s search engine spelling corrector. As the years went by, Shazeer shifted his focus toward deep learning and natural language processing (NLP), joining the elite Google Brain research division. He became fascinated by the constraints of language processing, famously noting that because text is immensely data-dense compared to images, it represents the ideal frontier for computation and true artificial intelligence. Despite his monumental breakthrough for LLMs, friction began to grow inside Google. In the late 2010s, Shazeer and his colleague Daniel De Freitas built a highly advanced conversational chatbot named Meena (later evolving into LaMDA). At the time, Shazeer was convinced that conversational AI was the future and aggressively pushed to release it to the public. Google executives, however, were bound by corporate caution and fears of reputational risk, and hence, repeatedly kept blocking the release. The corporate roadblocks instilled frustration in Shazeer and De Freitas, who eventually chose to walk away from Google in 2021. In 2022, OpenAI released ChatGPT to the world as a conversational AI, proving Shazeer was right. Nonetheless, Shazeer and De Frietas founded Character.AI as a startup that allowed users to create and interact with highly expressive, customisable AI personalities, ranging from fictional characters to historical figures. Character.AI exploded in popularity, achieving a valuation of $1 billion during its 2023 Series A funding round. Shazeer was born in Philadelphia in 1976 and was raised in an environment that deeply valued academic and intellectual pursuits. His family history is varied, with his grandparents having fled the Nazi Holocaust to the Soviet Union and later spending time in Israel before immigrating to the United States. Shazeer’s father, Dov Shazeer, was a multilingual math teacher who later transitioned into engineering, while his mother was a dedicated full-time homemaker. For Shazeer, the environment allowed him to develop an engineering mindset. Hence, it is unsurprising to see Shazeer developing an affinity for mathematics. He attended Swampscott High School in Massachusetts, where he earned a spot on the USA team for the 1994 International Mathematical Olympiad (IMO) – where he brought home a gold medal. He then attended Duke University from 1994 to 1998 on the prestigious Angier B. Duke Memorial Scholarship, where he graduated with a double major in mathematics and computer science, all while consistently winning prizes in the elite Putnam Mathematical Competition. Shazeer briefly entered a graduate program at UC Berkeley but the internet boom pulled him directly into industry before he could finish the degree.