As my narrowness talk accumulated 10k views in 3 days, I feel like clarifying a few behind-the-scene facts.
First of all, thank you for the attention! In this world of endless digital attention grabbers, I feel really lucky, largely undeserved, but also quite elated that some of you spent an hour (or 5 mins, 10 mins…however long it took for you to close that window) listening to me.
Here I am providing a timeline of life events leading up to, and after the talk.
- From March to August 2020 I was actively interviewing for a research scientist job – one that would allow me to keep doing what I was doing in Uber AI Labs, publishing ML papers.
- At that time, my scholar profile was a little bit under 1000 citations; I had NeurIPS x4, ICLR x2, ICML x1.
- I interviewed at Google Brain, FAIR, OpenAI, and Nvidia. All rejections. No offer.
- Out of desperation I also interviewed Facebook and Twitter for more applied ML roles, and some smaller companies for management roles, but evidently my credentials were not aligned enough (“You’ve never shipped real-world ML products” / “You’ve never had direct reports”), or perhaps they could tell that my heart was not in the right place.
- Each of the interviews was long and painful, on an average spanning over two months, and for the first time in the company’s history, entirely virtual.
- The “interview questions” I listed in the talk are real.
- Except for Google Brain, whose recruiter called me and explained in person that they were going through an (understandably) uncertain time, and hence would not be able to extend me an offer at that moment, every other place gave me a plain rejection letter, sent from a recuiter who clearly copied from a template. Note: I tend to believe most people have good intentions, but the execution is often beyond the intention’s reach. So I am not imposing any judgements.
- My personal relationship also ended abruptly then, making me think if this was altogether a sign from the Universe.
- ML Collective was incorporated August 2020.
- I stopped looking for a job and went full-time running MLC, unpaid, for a year.
- I continued publishing and participating at ML conferences, as if I had a RS job. Thank god virtual conference registrations are cheap.
- MLC started from an team of friends who agreed to continue meeting every week as if we were still a lab, and a reading group which I tried my best to not let die.
- In October, I perhaps gained some recognition from MLC’s growth, and was invited to give a talk at Google. I was told I could talk about anything I wanted. I decided on the topic almost the last minute, and was totally stressed out about it the night before.
- After the talk, over 100 emails landed in my inbox from Googlers who saw the talk. I was Goolge-famous for a minute.
- A few weeks later the amazing Doug Eck reached out, asking about my interest in considering Google Brain again.
- I wasn’t just handed a job; I had to go through another extensive interview cycle, including giving a research talk, many coding and research qualifiers, spanning again over a month. I remember thinking how stupid it would be if it were yet another rejection.
- In April 2021, I officially joined Brain, under a special contract that allows me to continue running ML Collective.
You shouldn’t trust any conclusions drawn from a single data point. But if I were to write down some insights solely for the future me to read when she ever feels lost again:
- Failures are the most valuable, if not only, opportunity where one gets to reflect and learn about oneself.
- When you encounter a failure, think: is it a “conventional failure” (one that the majority in soceity consider a failure, for you and of you), or is it “your failure” (one that you, deep down, know that you could do better)? Only the latter requires making changes to yourself.
- “Just be so awesome that no one can ignore you.” – Jason Yosinski, rephrased from comedian Steve Martin.