Ask HN: Do I Need a PhD?

[throwaway_am]

Hello *

I'm a Physicist with more than 6 years of experience in Data Science/Machine Learning in industrial/business environments. I like my job, but sometimes I miss the old days when I was doing research during my master's degree. Therefore, I've started applying to private companies that do ML research (OpenAI, Meta, DeepMind, NVIDIA, etc.). However, even if not asked explicitly, it seems impossible to be accepted in any of these companies without a PhD in a relevant field. My CV is not bad -I would say above average- but I never did a PhD.

The point is that recently, I received an opportunity to pursue a PhD in Machine Learning/Deep Learning, and now I'm wondering if it's a good move to accept the opportunity and leave my current job. While I'm not interested in an academic career in university (I don't see myself moving from one university to another for the rest of my life), it seems that a PhD is necessary to work in ML research.

What do you recommend? Should I leave my current job and take the risk of pursuing a PhD, or should I continue to apply to these companies and wait for an opportunity to arise?

Thanks for your help!

[PaulHoule]

A FAANG company that has its choice of who it wants to hire can require a PhD, other companies that need talent but will be blown off entirely by many candidates because it isn't FAANG will hire who it can.

My take is that until relatively recently, CS was a place where people got further without PhDs than other fields. As late as the mid-2000s, Carl Lagoze was a PI at Cornell on a very large project and taught classes before he got his PhD. There is no way you'd find somebody in physics like that because for any physics job there are 20+ PhDs who want it.

A PhD, however, is (potentially) the right training for what you are trying to do. I say "potentially" because you can easily get diverted into "me too" research that is citeable and fundable right now; it's not so much that involvement with this develops your skills in the wrong direction, but it definitely can develop your attitudes in the wrong direction for industrial research.

[johndoe0815]

You should consider that by the time you finish your PhD (in 3-6 years), the demand for experts in ML/DL will have gone down significantly - either because the next AI Winter has finally arrived or there is simply an oversupply of experts in this area since currently the majority of CS students wants to work in AI/ML/DL.

In contrast, there are a number of open PhD student positions in systems here in Germany for which it is extremely difficult to find appropriate candidates. The few students interested in systems topics receive excellent offers from they industry and leave academia after graduating with a master (and sometimes even bachelor) degree.

[jstx1]

> I've started applying to private companies that do ML research (OpenAI, Meta, DeepMind, NVIDIA, etc.)

If you want to do research at a top company, yes, you probably need one. Or at the very least you need to somehow compete with other people who have relevant PhDs. It's still kind of risky because who knows where the goalposts will be by the time you finish your PhD.

[throwawaysalome]

He's all right. He's a goodfella, one of us. -- Goodfellas (1990)

Consider it from the perspective of the hiring manager who also has a PhD. Why spend time vetting someone without a PhD when your CV inbox is glutted with vouchsafed PhDs.

At your age, you won't get into a name PhD program. The prestige of a FAANG skunkworks job, even if you could get it, doesn't sound worth the six-year slog of unpaid and politically unpleasant work.