Silicon Valley AI talent scramble: OpenAI opens up to 1 million knives, Google large stock incentives
PUBLISHED
2024-01-22
It’s been nearly a year since ChatGPT brought on the craze, and the competition on big models remains fierce for both Silicon Valley tech majors and startups. But underneath the rapid iteration of models and products – tapping into talent – is the dark side of this competition.
According to The Information, citing people familiar with the matter, Google has set up a special stock compensation pool in the hope of countering OpenAI’s offensive to scoop up people at any cost and retaining top AI researchers. Within DeepMind, Google’s core division, a number of researchers working for Gemini have received restricted stock ranging from hundreds of thousands to millions of dollars.
According to Forbes statistics, in the past five years, AI researchers who left Google have founded at least seven artificial intelligence-related startups –
Ilya Sutskever, co-founder of OpenAI
Dario Amodei, Christopher Olah, co-founders of Anthropic
Aidan Gomez, Nick Frosst, founders of Cohere (chatbots for the enterprise)
Noam Shazeer, Daniel De Freitas, co-founders of Character.AI (customizable chatbots)
Niki Parmar and Ashish Vaswani, co-founders of Adept
Jakob Uszkoreit, co-founder of Inceptive (AI life sciences company)
Llion Jones, founder of Sakana AI
Last August, with the announcement of Llion Jones’ departure to start his own business, all eight AI scientists who published the famous paper “Attention Is All You Need” and proposed the Transformer architecture have left Google.
Although this equity incentive is a bit late, but the need to retain the AI team can not be delayed.
Many times to poach Google’s OpenAI
After the big model competition, OpenAI can be said to be the “first stop” for many Google employees who left the company.
According to data statistics company Live Data Technologies data, since December 2022 to date, Google has been OpenAI’s largest source of talent inflow, from Google and DeepMind team members accounted for 7.5% of the total inflow ratio, much higher than the second place Stripe and the third place Meta.
Live Data Technologies
Google’s AI talent also contributed a lot to the birth of ChatGPT. According to The information, before ChatGPT was released, OpenAI poached five Google researchers to do pre-release optimization for the product.
openai
Since last October, OpenAI has poached at least two mid-level executives from Google’s Gemini team who manage dozens of employees. One of them is reportedly Jiahui Yu, who is in charge of multimodal model development, and the other is Amelia Glaese, Google’s human data lead, who oversees how Google uses people’s feedback to improve its models.
Of course, OpenAI has been actively courting people both explicitly and implicitly, and Google has poached far more than that. Other researchers who worked on the Gemini project, including Jonathan Uesato, Maja Tr bacz, Keren Gu-Lemberg, and Tao Wang, have also switched to OpenAI since November of last year, according to publicly available LinkedIn profiles and people familiar with the matter.
However, for the most heavyweight talent Google has sent to OpenAI, it’s none other than OpenAI co-founder and chief scientist Ilya Sutskever.
Ynetnews
In 2013, Sutskever joined Google Brain, during which time he created sequence-to-sequence learning algorithms with two other coworkers and participated in the development of the open source deep learning framework TensorFlow. It wasn’t until 2015 that Ilya was poached by Musk for OpenAI.
Musk later went on record as saying that the poaching was one of the toughest recruitment battles he’d ever experienced, and that he even went so far as to make a complete break with Google’s founder in order to poach Ilya.
Ilya did not disappoint Musk’s expectations. Since joining OpenAI, Ilya has played a key role in the development of large-scale language models, as well as promoting the development of GPT-2, GPT-3 and the Venn diagram model DALL-E, and some people even believe that he should be called the “father of ChatGPT”.
However, OpenAI’s tradition of cutting corners starts with current CEO Sam Altman, who published two articles, “How to hire” and “Employee Retention”, back in 2013, and said that founders should spend one-third to one-half of their time on hiring.
Sam Altman also says directly in the article that the really good people are usually not actively looking for work, so you need to go digging for the talent you are looking for. He also quotes a friend:
“Poaching is like fighting for your life in a Silicon Valley relationship.”
OpenAI is even more relentless when it comes to salaries. According to data published in April 2023 by Rora, an overseas company that helps techies negotiate salaries, OpenAI is the most generous employer when it comes to compensation for employees working in AI, averaging up to $865,000 per employee per year.
AIM
According to this figure, OpenAI offers its employees a salary that is 25% higher than Google Brian and 92% higher than Apple. So in addition to Google, there is also a lot of talent exodus from old tech companies to OpenAI.
Last July, former FAIR (Facebook AI Research Institute) researcher Noam Brown announced that he was joining OpenAI, saying that he hoped to use the AI self-pairing and reasoning in games such as poker and Diplomacy, which he had been working on, for the optimization of large language models.
In addition, according to Bloomberg, Altman and Jony Ive recently teamed up to “pry away” Apple’s VP of product design, Tang Tan, who was head of design for the iPhone and Apple Watch, and who will be joining Jony Ive’s startup, LoveFrom, in February. In February, he will join Jony Ive’s startup LoveFrom to develop hardware with OpenAI.
A seller’s market moment for AI talent
But Silicon Valley talent moves so fast that even the big Silicon Valley tech companies, which are notoriously well-paid, can’t always keep the top talent.
Some leave to start their own businesses, like siblings Dario Amodei and Daniela Amode, who were OpenAI’s vice president of research and development and vice president of security policy. After OpenAI received a large investment from Microsoft, they decided that the company had strayed from its original vision and went out on their own to start Anthropic, which is now the second most influential startup in the industry after OpenAI.
Timothee Lacroix and Guillaume Lample, former Meta employees and authors of the Llama paper, moved from “big models” to “small models” by co-founding the AI startup MistralAI, a star startup that recently received its latest round of funding at a valuation of $2 billion.
The big tech companies are constrained by their original business and can’t turn quickly, which can also be a problem. On the one hand, it’s true that big companies in Silicon Valley are more cautious because their every move is under the scrutiny of the media and the public, and are far less free and flexible than smaller companies. At the same time, the more complex internal hierarchy of large companies, coupled with the already mature and profitable business, leads to large companies in the capture and response to market trends much less quickly than only the core team of AI startups.
Some Google Brian employees have revealed that Google’s internal attitude towards AI was very cautious in the past. As early as 2021, Google announced the creation of a large language model based on Transformer that can understand and generate conversations with humans, but due to the fear that the technology is not accurate enough, Google’s internal hesitation to launch a chatbot is the main reason why Google can’t keep employees who want to fulfill their ambitions.
Just days ago, Diane Hirsh Theriault, a senior engineer at the company, criticized the company’s executives for having “no vision” in a long post on LinkedIn.
LinkedIn
Of course, some employees are forced to leave because they can’t stand the company’s internal competition.
Among the 14 authors of the original research paper on Meta’s Llama large model in February last year, more than half of the AI scientists have left Meta. according to foreign media analysis, the main reason for the employees to leave is the uneven distribution of limited arithmetic resources within the company. The company’s teams in Paris and the United States have been independently developing the Llama large model and the OPT model, respectively, and the conflict between the two teams ultimately led to the departure of Antoine Bordes, the head of the Paris team, to join the military AI company, Helsing.
In addition, having an unreliable leader at a company can accelerate employee departures, such as Stability AI, which gained notoriety for Stable Diffusion.
Last June, more than 30 people, including former employees and investors, reported that CEO Emad Mostaque was guilty of several offenses, including massive non-payment of wages to employees and “stealing” their research. Under Emad Mostaque’s leadership, more than a dozen Stability AI executives have left the company, and Forbes even predicted Stability AI’s demise in its 2024 AI industry outlook.
It’s a seller’s market moment for AI talent in Silicon Valley.
Meta’s CEO Zuckerberg expressed a similar sentiment in a recent interview with The Verge: the battle for AI talent is as fierce as ever, with every company competing for a very small number of researchers and engineers. Those with specialized skills can get jaw-dropping pay packages, and Zuckerberg himself is often directly involved in the hiring process.
That’s the reason that after OpenAI’s management change, Microsoft cut its own employees’ bonuses and stock while also biting the bullet and recruiting those who left OpenAI en masse at what would have been high salaries. Also waiting in the wings early were Anthropic, Adept, Cohere, Replit, and a host of other AI startups.
Retaining employees with a sense of mission and growth
In addition to the reasons mentioned above, the lack of non-compete agreements in Silicon Valley is a big reason why employees can move around so freely.
On the one hand, California, where Silicon Valley is located, legislated against non-compete agreements in 1872, and in January 2023, the Federal Trade Commission stated that it intends to prohibit non-compete clauses between employers and employees nationwide. On the other hand, tech companies have a high level of confidentiality within their ranks, and employees who steal trade secrets are subject to legal sanctions.
Historically, the flow of talent in Silicon Valley has been freer since then, although the big companies in Silicon Valley used to agree with each other not to cut corners but were sued by the U.S. Department of Justice in the name of monopolization.
So what can a company do to retain its core employees?
Sam Altman’s thinking may be worth learning, after all, before he was kicked out of the company by the board of directors at the time, had triggered several employees to jointly request a collective resignation and he went forward and backward.
In his article, “Retaining Employees,” he suggests three main approaches:
Sense of Mission
According to Altman, if employees work because they believe in the company’s mission, they are less likely to be lured by higher salaries offered by other companies and are less likely to be eager to leave to start their own business.
Rapid growth
Rapid business growth means that everyone is constantly faced with new challenges and can provide employees with an expectation of wealth freedom by granting them equity (something that can never be achieved with a salary alone).
A great work environment
Altman believes in defining cultural values early on and maintaining them over time. At the same time, never compromise in the first ten hires because great people always want to work with other great people.
In response to OpenAI’s onslaught, Google is now focusing on retaining employees through stock and salary increases, the effects of which are yet to be seen, but the added pressure on the company’s payroll expenses is real. in the third quarter of 2023, Google’s unallocated corporate costs, the vast majority of which come from DeepMind, jumped nearly 40% to $1.6 billion. At the same time, Google has been restricting hiring over the past year and recently laid off more than 1,000 employees, with the savings going back to make up for the rising cost of AI talent.
The good news is that Google, becoming generous, has indeed been poaching employees from OpenAI in reverse as well. An AI engineer named Matt Wiethoff said on LinkedIn that he jumped ship to Google starting last October, after leading the development of a popular ChatGPT feature called Code Interpreter that automates data analysis.
However, it remains to be tested whether this strategy will be effective in the long term.
