Here's what it takes to work at Google DeepMind - a London startup no one has ever left
Today some of the smartest people in the world are queuing up to work at DeepMind, according to an article by Celemency Burton-Hill in The Guardian in February. Interestingly, the same article states that no one has ever left DeepMind, which has created a series of algorithms that can learn for themselves and beat the best humans at games like Go and "Space Invaders."
Based in up-and-coming King's Cross, DeepMind now employs around 250 people. However, as Burton-Hill points out, getting a job there is far from easy.
Fortunately, a number of Quora Q&As offer an insight into "What does it take to work at Google DeepMind?" and "What is it like to work at Google DeepMind?"
Responding to the first question, Imperial College London computing graduate and Google DeepMind engineer Matthew Lai wrote that most of what you read about Google's hiring process doesn't apply to DeepMind.
"Their recruiting is still separate from rest of Google," said Lai, adding that DeepMind hires research scientists, research engineers, and pure software engineers.
Lai, himself a research engineer, said he was interviewed for a total of about eight hours by a software engineer, a senior research scientist, and one of the DeepMind founders. Before getting his job offer, Lai said he also had a "Google Hangout quiz" that featured questions on machine learning, stats, and maths.
"If you want to join as a research scientist, you'll need a PhD and probably a few years of machine learning research experience," said Lai. "It sounds quite competitive, and all the research scientists I have talked to have very impressive credentials and experiences from either academia or other industrial research labs.
"If you want to join as a research engineer (still in research, but a bit less theory and more practice), a PhD is not required, though I believe most people still have at least MSc, and significant machine learning experience."
Lai said interviews for research engineer positions include two hours worth of questions (covering subjects such as statistics and maths) and some algorithmic coding. "The focus is more on the practical side, but you definitely still need to know your machine learning as well," he said.
Gary Wang, an academic at the University of Alberta with expertise in artificial intelligence, wrote on Quora that it helps if you go to certain target schools. "For example, here at University of Alberta, plenty of alumnis both in the masters and PhD level get jobs at DeepMind, due to a very very (arguably the best?) reinforcement learning group in the world, with professor Rich Sutton, Csaba, etc," he writes. In another Quora answer Wang mentions Stanford, Oxford, University College London and Carnegie Mellon University.
Not everyone on Quora needs help getting a job at DeepMind. One anonymous user appeared to be having difficulty deciding whether to accept a job offer from DeepMind, explaining that he'd also received offers from Google Brain, a deep learning research project at Google, and Facebook AI Research (FAIR), which is a group of people within Facebook who are "committed to driving the field of machine intelligence," according to FAIR's website.
Kabir Chhabra, a computer science student in Delhi, India, responded to the anonymous user with: "Personally, Google Deepmind sounds like the kind of mythical place where superhuman convene to discuss the matters of AI and bless us with their outcomes. Facebook AI Research is new, but they've poached some of the best people from around the earth, and have crazy amounts of funding to splurge on you. As I said, you can't really go wrong whatever you choose to go with!"
DeepMind also provides some of its own advice on its website. On the "Join Us" section of its website, DeepMind writes:
While there are no specific roles advertised on the website, DeepMind states that it's "always on the lookout for exceptional people" and encourages prospective applicants to email them with a CV and any other personal information that may be relevant.
Google declined to comment.