Getting a sought-after job at a top AI lab may require a surprisingly old-fashioned approach, according to a Google DeepMind distinguished engineer.
“Work like a dog,” Vladimir Feinberg, a researcher who leads Gemini pretraining at Google’s AI lab, wrote in a recent blog post titled “How to Land a Frontier Lab Job.”
Feinberg explained that securing a job at a frontier lab is especially challenging, given the fierce competition.
“There is always a vanguard of elite college students (both undergrads and PhDs) who do ML research in top-tier conferences, math and programming competitions, and already have connections to these labs through older classmates or friends,” Feinberg said.
Years ago, that same pool of talent may have been recruited by Wall Street stalwarts like Citadel and Jane Street. Today, Feinberg said, many of those top students are instead competing for coveted roles at AI companies like OpenAI, Anthropic, and Google DeepMind.
“The reason this cohort is so successful is because the underlying traits are highly predictive of success,” Feinberg wrote, pointing to three key qualities: “intent,” “mathematical maturity,” and “grit.”
“This brings me to the advice I’d give myself if I was entering college today: do everything you can to join that cohort mentioned above,” he wrote, recommending that aspiring AI researchers take “difficult, proof-based classes,” “code, obviously,” and use AI tools “for what you already know how to do, only, but aggressively so.”
That means putting in long hours outside the classroom and sacrificing nights and weekends to develop the skills needed to compete for jobs at leading AI labs, according to Feinberg.
“There is no substitute to the above for achieving mathematical maturity, which is essential, but setting this aside, the most obvious way to get hired by a lab is to demonstrate that you have a specific skill that the lab requires,” Feinberg wrote.
He acknowledged that breaking into an AI lab can feel like a catch-22.
“The way out of this is to work at the edges of where frontier labs operate: they spend their time creating LLMs. What do LLMs require to run, and what are the touchpoints for their outputs? This is the direction where frontier labs expand their scope, and thus are the few specific areas that don’t require training LLMs, but are nonetheless essential to the business,” he wrote.
Feinberg told Business Insider he has one additional piece of general career advice — “be the kind of coworker that people would want to see succeed.”
“Identify opportunities for your team’s complementary skillsets to shine, credit collaborators concretely to their leadership, and identify projects where your success leads to that of others,” he added.
In an episode of The Peterman Pod released on June 15, Feinberg said that after he published the blog post, he heard from people at Anthropic and OpenAI who said they agreed with his advice.
“There’s maybe differences in like business strategy and the set of offerings that’s a function of the specialties of the labs and the kind of different customers that the labs could have,” he said. “But I would say that there’s quite a lot of overlap between the labs in terms of what people look for.”
Asked by podcast host Ryan Peterman whether advances in AI could diminish the value of research work, just as some fear it will for software engineering, Feinberg said he believes the “research skill set is going to become increasingly important.”
“Thinking about how do I construct systems around these LLMs to do my job more effectively — that’s going to be the thing that sets you apart in the future,” Feinberg said. “And I think that’s true no matter what you’re going to be doing.”
Read the full article here
