This as-told-to essay is based on a conversation with 41-year-old Devi Parikh, who lives in San Francisco. The following has been edited for length and clarity.
The seed of my passion for AI was planted in the early 2000s when I studied electrical and computer engineering at college. I was exposed to a type of machine learning called pattern recognition.
In 2009, I completed a Ph.D. in computer vision at Carnegie Mellon — well before the current excitement around LLMs and generative AI. But we had the same goal: make machines more intelligent.
Next, I moved into research and teaching roles, and in 2016, I spent a year as a research scientist at Facebook AI Research, or FAIR. Following that, I’d spend my springs and summers at FAIR in Menlo Park, California, and my falls teaching computer vision at Georgia Tech.
Over time, I enjoyed Meta more than my professorship, and I transitioned to a full-time role in 2021, eventually becoming a senior director of GenAI.
In 2024, I left Meta to start an AI company called Yutori, alongside my husband and our friend.
Here’s what I’ve learned about getting into and succeeding in AI after over 15 years in the industry.
1) Don’t assume you need a Ph.D. to do cutting-edge AI work
Professor and research scientist roles in AI might list a Ph.D. as a requirement, but there are other cutting-edge jobs in this space.
There are good reasons to do a Ph.D, like if you want to work in academia or explore certain ideas. But if your end goal is doing interesting AI work and learning how the sausage is made, you could spend those five to six years at startups or big labs instead.
You can also try side projects, making use of open source code and online communities to get your hands dirty.
If you keep putting in the time and effort to whatever you’re doing, you’ll be able to stand out, and you’ll also have learned a bunch of skills along the way.
I think the perception that a Ph.D is necessary in this industry has changed over time. We don’t take them into consideration much when hiring at Yutori, where we’re trying to build AI agents that can help people with digital chores, like looking for apartments or buying headphones.
Courtesy of Yutori
Instead, we look for people with relevant experience, such as in training models, and how candidates perform in technical interviews involving coding problems and system design questions.
2) Keep your professional identity flexible
Between 2011 and 2013, there was a “deep learning wave,” when the AI community began to realize the effectiveness of deep neural networks.
Some fellow researchers tied their identity to the tools they had worked with and were hesitant to transition to deep models, even though it was clear they worked much better for the problems we were addressing.
This field evolves rapidly, and if evidence tells you new tools work better, don’t hold onto your past tool set. Holding on to your professional identity, such as by seeing yourself only as an academic, can also be detrimental.
I also learned not to hold on to research areas. I worked on computer vision during my Ph.D, then multimodal problems, and later generative models for images and videos. At the time, I didn’t know ChatGPT was coming, and that generative AI would suddenly become a high priority in tech. If I’d held onto my identity as a computer vision researcher without exploring these other things, I would’ve missed out on opportunities.
3) Pursue your genuine interests, not what you think you should do
On paper, my job at Meta was amazing, and you probably wouldn’t leave it to start a company if you were being strategic about moving up in your career, and knew the success rate of startups.
It may be unclear whether an opportunity is the right move strategically, but I find it easy to put time and effort into things I think will be fun, and produce better quality work that gets recognized.
4) Follow through on ideas
Seeing things through to the end — 100%, not 95% — may be the single most important thing that’s helped me stand out and achieve what I have.
For example, during the COVID-19 pandemic, I started a series on YouTube called “Humans of AI,” where I interviewed around 20 AI researchers in my network about their daily habits, strengths, and insecurities. I thought seeing the human side of the AI researchers we put on a pedestal would show folks in the community they could have a similar level of impact.
People loved it, and it made me more visible. I’ve met people at conferences who might not have known about my research, but saw the series.
Many people are excited 20 or 30% into the execution of their ideas, then their interest tapers off, leaving behind a bunch of unfinished projects. If you haven’t seen something through to the end, it can’t have its impact or lead you to the next thing.
If there’s something you’d like to do, just go do it, instead of overanalyzing and not taking steps forward.
Do you have a story to share about building a career in AI? Contact this reporter at [email protected].
