AI search engines stop using Google

According to Parmy Olson, all eight authors of a groundbreaking AI research paper have departed from Google and embarked on their individual entrepreneurial endeavors.

It is quite fitting that one of Google's most significant creations, which would later prove to be a source of trouble for the company, was originally conceived during a casual lunchtime conversation.

During 2017, a group of researchers situated at Alphabet's headquarters in Mountain View, California engaged in a lunchtime discussion regarding the enhancement of text generation efficiency by computers. Little did they anticipate the immense significance of their subsequent discoveries. Over the course of the following five months, they conducted experiments and documented their findings in a research paper titled "Attention is All You Need." The outcome proved to be a groundbreaking advancement in the field of AI.

The transformative work of the eight authors resulted in the creation of the Transformer—a revolutionary system enabling machines to generate an array of data, including humanlike text, images, DNA sequences, and various other forms, with unprecedented efficiency. Their research paper garnered remarkable recognition, accumulating over 80,000 citations by fellow scholars in the field. Furthermore, the AI architecture they developed became the foundation for OpenAI's ChatGPT (where "T" represents Transformer), as well as other innovative tools like Midjourney, designed for image generation and beyond.

The act of Google sharing this groundbreaking discovery with the public was not an uncommon practice among tech companies. Often, they open source novel techniques to gather feedback, attract skilled individuals, and foster a supportive community. However, Google itself did not immediately utilize the newly developed technology. Instead, the system remained relatively dormant for years while the company focused on the broader challenge of transforming its cutting-edge research into practical services. In the meantime, OpenAI seized the opportunity and leveraged Google's own invention to launch a significant threat to the search giant—an occurrence not witnessed in years. Despite Google's wealth of talent and innovation, it was competing firms that ultimately capitalized on this major breakthrough.

Furthermore, the researchers who co-authored the 2017 paper did not envision a long-term future at Google either. In fact, all of them have since departed from the company and embarked on their individual paths. Among their ventures are startups like Cohere, specializing in enterprise software, and Character.ai, founded by Noam Shazeer, who held the distinction of being the longest-serving member of the group and was regarded as an AI luminary within Google. Collectively, their businesses now hold an estimated value of around $4.1 billion according to valuations provided by research firm Pitchbook and price-tracking site CoinMarketCap. Undoubtedly, they have become prominent figures in the realm of AI within Silicon Valley, earning the status of AI royalty.

This week, Llion Jones, the final remaining author among the eight, officially announced his departure from Google to embark on his entrepreneurial journey. Reflecting on the rapid growth and impact of the technology he co-created, Jones expressed a sense of surrealism. "I've only recently felt... famous?," he confided to me. No one recognizes my face or name, but it takes merely five seconds to explain: 'I was part of the team that developed the 'T' in ChatGPT.'" The experience has brought forth a unique blend of recognition and anonymity for Jones, as he witnesses the widespread influence of the technology he contributed to.

It seems odd that Jones' rise to fame stemmed from activities outside of Google. Where did the business make a mistake?

There is a clear issue that stands out: scale. According to an estimate by Glass.ai, an AI firm that analyzed LinkedIn profiles to identify AI employees at major tech companies, Google has a staggering workforce of around 140,000, with 7,133 dedicated to AI. In contrast, OpenAI, which ignited an AI arms race, operates with a significantly smaller team. In 2023, they had approximately 375 staff members, out of which only about 150 were AI researchers.

During the development of the Transformer, numerous former scientists and engineers revealed that Google's immense size posed a challenge. They mentioned that the hierarchical structure necessitated multiple layers of management approval for ideas. Additionally, researchers at Google Brain, a prominent AI division, faced a lack of distinct strategic direction, leading many to become overly concerned with personal career growth and their prominence in research publications.

Google's criteria for transforming ideas into viable products were notably stringent. Illia Polosukhin, who initially discussed ideas with fellow researchers Ashish Vaswani and Jakob Uszkoreit at the Google canteen, remarked that Google would only pursue concepts that had the potential to become billion-dollar businesses. However, establishing a billion-dollar enterprise entails continuous iteration and encountering numerous setbacks along the way, which Google didn't always have the patience for.

According to a Google spokeswoman, the company takes pride in its pioneering and groundbreaking contributions to the field of Transformers. They expressed enthusiasm about the vibrant AI ecosystem that has emerged as a result, which includes collaborations with new partners. The spokeswoman also acknowledged the bittersweet reality that these opportunities allow their researchers to further advance their work beyond Google's confines.

In a sense, the company fell victim to its own achievements. With renowned AI scientists like Geoffrey Hinton as part of its workforce and an early adoption of advanced AI methods for text processing as early as 2017, there was a prevailing mindset among many researchers of "If it ain't broke, don't fix it."