📚How I went from finance → AI
How I went from dead-end financial consulting and no technical skills to machine learning engineering in 5 years, and then to AI research 3 years later.
My guiding principle was to maximize my usefulness - I never planned farther than 6 months ahead. If I'd seen this roadmap in 2012 I probably would've given up immediately and started that pub crawl business.
| Time | Status | Emotions | Do differently |
|---|---|---|---|
| 2011 Q2 | Graduated with econ degree | Aghast | Things don't just "work out" |
| 2011 Q3 – 2012 Q4 | Job 1: financial consulting | Misery | Everything |
| 2013 Q1 | Web Development curriculum (step 1) | Elation | |
| 2013 Q2 – Q4 | Dev Bootcamp | Obsession | |
| 2013 Q4 | Job 2: Web development | Relief | Find mentors |
| 2014 – 2015 H1 | Stagnation | Frustration | Everything |
| 2015 H2 | Dipping my toes in CS | Revelation | |
| 2015 Q4 – 2017 Q4 | Job 3: Fullstack Engineer | ||
| 2016 Q1 | First attempt at CS Curriculum | Frustration | Ignore unhelpful degrees |
| 2016 Q2 – 2019 H1 (3 years) | CS Curriculum (step 2) | Epiphany | Project sequencing |
| 2016 H2 | Dipping my toes in ML | Curiosity | Sabbatical sooner |
| 2017 H1 – 2019 H1 (2 years) | ML Curriculum (step 3) | Narrow scope | |
| 2017 Q4 | ML sabbatical | Frustration | Better curriculum |
| 2017 Q4 – 2020 Q4 | Job 4: ML Engineer | Imposter | Ask for help |
| 2019 H2 – 2020 H1 | NLP Systems Curriculum | Overconfident | Focus on other bottlenecks |
| 2020 Q1 | CS294 Deep Unsupervised Learning | Challenged | |
| 2020 Q4 – 2021 Q2 | Job 5: ML Researcher | Amazed | Nothing! |
I loved my job as an AI researcher. Research is simply learning the answers to questions that nobody has asked before, so the learning skills I’ve accumulated served me well. But I had no choice but to quit: the Archive had broken up during the pandemic, and I couldn’t stop thinking about starting a community that would last for 50 years instead of 5.
I usually encourage people to start programming if they love to build and could imagine themselves being a programmer for 5+ years.
There are many folks with business skills that want to “learn how to talk to engineers”, learn how to build their own MVPs, or otherwise don’t plan to be a full-time programmer for the bulk of their career. I usually discourage these folks from going to a bootcamp or taking intro to programming courses. Instead, I’d bet on the no-code movement. Here’s a curriculum I prepared for a friend: