After ChatGPT's blockbuster, Google sounded the internal "red alert" for the first time, and there were many heroes who turned against each other and jumped to OpenAI. Previously, foreign media reported that OpenAI has poached at least a dozen AI experts from Google in a few months since ChatGPT was launched in November last year, including Jacob Devlin, the father of BERT and the industry's top expert.
Somewhat surprisingly, however, Devlin returned to Google just a few months later, as if the story had not happened for months. As a bystander, in addition to being surprised by this series of changes, it is inevitable to wonder if there is really no undercurrent underneath the calm wind?
Devlin is his own thing
When it comes to the name of Devlin, it can be said that everyone knows everyone in the AI community and even the entire technology circle. Before joining Google, he worked as a software engineer at Microsoft for more than three years, officially joined Google in 2017, and the following year he proposed the concept of BERT and published related papers with colleagues. BERT is an acronym for Bidirectional Encoder Representations from Transformers, which translates to "Bidirectional Encoder Representation Technology Based on Converter Mechanism". BERT was derived from the transformer architecture, and like the GPT series in the following years, this paper and the research it contains have become part of the language model of Google and OpenAI. Later, many large model companies have borrowed or learned from BERT and GPT at the beginning. So if BERT is the beginning, then Devlin can be called a generation of grandmasters.
As soon as BERT was announced, Google used it to better understand the semantics of user search statements. In 2020, a literature survey and analysis concluded: "In the past year or so, BERT has become a ubiquitous baseline in NLP experiments. "Counting the analysis and research on BERT's improved model, more than 150 publications were published in just over a year.
Regarding the reason for this return to Google, neither the official nor Devlin himself has said. According to some people familiar with the matter, Devlin will work closely with Bard's department after his return, and Devlin's departure is also related to Bard. Bard link: https://bard.google.com/?hl=en
Protests and job-hopping
Google announced Bard in February, followed the official launch of GPT-4 in March. However, news soon emerged that Bard's team was using information from ShareGPT to train its own product, and Google acquiesced. ShareGPT is a Chrome extension that launched last December to capture and share conversations people have in ChatGPT. According to OpenAI's user protocol, ChatGPT's training results cannot be used commercially. Obviously, Google's operation clearly violated the relevant regulations. Therefore, after the news was exposed, it quickly attracted many practitioners' verbal criticism, on the one hand, because Google has always prided itself as a global AI industry leader, but it has so blatantly ignored user agreements, on the other hand, it is also out of disdain for Google's lazy behavior. ShareGPT link: https://sharegpt.com
Later, foreign media reported that as early as before this news was exposed, Devlin expressed protest to the leadership of Google and its parent company Alphabet, making it clear that this violation of the user agreement would inevitably attract criticism from the outside world, and would also lead to Bard's answer being similar to ChatGPT, or even the same, unable to reflect Google's own differentiated advantages. After the protest, Devlin jumped to OpenAI, and it is unknown how Google reacted to Devlin's protest.
People in the AI world are money
Generally speaking, personnel changes between companies are normal. But Google is different from OpenAI, not only because they are star AI companies in the world's attention, once a little wind and grass will immediately attract attention from the outside world, more importantly, they are also competitors, and the trade-off of personnel changes will inevitably affect their respective products and even the status of the company.
It's worth noting that Devlin is just one of many companies from Google to OpenAI or others. In November last year, when OpenAI officially announced the launch of ChatGPT, there were several former Google employees in the online media acknowledgements. Faced with this situation, people can't help but sigh how did Google become the "Whampoa Military Academy" in the AI industry? It may also be that as some sources say, the formation of this kind of incense outside the wall is not overnight, nor is it limited to Google. Whether it's Devlin leaving Google for a few months and returning to Google, or OpenAI poaching people everywhere, you can see a phenomenon that the battle for AI talent is becoming increasingly hot.
With Devlin's return to Google, it was not surprising that it soon triggered online discussions on topics such as AI talent job-hopping, technology companies recruiting high-paid people, etc., some people agreed that some people eat melons, and some people said that they just wanted to wait for a wave of Bard's big update. Earlier, it was reported that Silicon Valley giants offered very attractive salary conditions for AI talents, and Google and Microsoft did not hesitate to spend thousands of dollars. Even domestic AI-related companies are also fighting for the head. A few months ago, Wang Huiwen, the number two figure in Meituan, released a "hero post", announcing the establishment of a light year away, researching domestic large models, personally contributing 50 million US dollars, and recruiting talents, dedicated to building a Chinese version of OpenAI, and clearly expressed his desire to recruit top R&D talents recognized by the industry. Of course, now people are more concerned about his condition, and the big model that does not know when it will be online. In addition, headhunters on various recruitment websites have also begun to cooperate intentionally or unintentionally this year to help set off a "recruitment wave" of AI talents. At present, BAT and other companies have a significant demand for positions such as "algorithm engineer" and "AIGC application expert", with monthly salaries of up to tens of thousands.
It is undeniable that money is indeed an indispensable part of the development of technology such as AI, but it is overkill, or impractical, to want to develop high technology just by burning money. At this time of unprecedented change, it is time to seek the unbiased golden mean. Although at present, "foreign affairs do not depend on US imperialism, and internal affairs do not depend on capital" is the habitual thinking of many people at present, is it really appropriate to dump such a big pot on external reasons?