Tracks Overview
The map. “Fellowship” is an umbrella word that tells you almost nothing — what actually matters is the track (the kind of work you do) and the program (the specific place you do it). This note is the tracks. For the places, see Programs Directory.
How to use this page
Read the one-liners. Notice which two or three make you lean in. Open those. Ignore the rest for now — you can always come back.
The whole map at a glance
| Track | You spend your days… | Lean if you like… | Open the note |
|---|---|---|---|
| Research Scientist | Asking new questions, designing experiments, writing papers | Math, ideas, “why does this work?” | Research Scientist |
| Research Engineer | Building the code & infra that makes research happen | Building, shipping, fast feedback loops | Research Engineer |
| Interpretability | Reverse-engineering what’s happening inside models | Puzzles, microscopes, “let’s look inside” | Interpretability |
| Alignment & AI Safety | Making powerful AI do what we intend, safely | Big stakes, careful thinking, the future | Alignment and AI Safety |
| Policy & Governance | Shaping rules, institutions, and incentives around AI | Writing, systems, persuasion, the real world | Policy and Governance |
| Applied & Product ML | Turning models into things people use | Impact you can see, end-to-end ownership | Applied and Product ML |
| Field-building, Ops & Comms | Growing the field — programs, community, communication | Organizing, teaching, connecting people | Field-building and Comms |
How they relate (the mental model)
Two families and a bridge
Picture a river.
On one bank: research — people producing new knowledge. Research Scientist, Interpretability, a lot of Alignment and AI Safety.
On the other bank: the world — people shaping how the tech lands. Policy and Governance, Field-building and Comms, a lot of Applied and Product ML.
The bridge is engineering. Research Engineer is the bridge on the research side; applied ML is the bridge on the product side. Engineering is how ideas become real, which is why it’s the most common entry point even for people who end up doing something else.
Almost nobody is purely one thing. The track is your center of gravity, not a cage.
The honest version of “which track”
A few things I wish someone had told me plainly:
- You don’t pick a track for life. You pick a track for your next application. People move between these constantly. I know research engineers who became scientists, scientists who went into policy, policy people who learned to code.
- The fastest-to-enter tracks for beginners are usually engineering and governance applied roles, because they reward demonstrable skill (a repo, a memo) over credentials.
- The hardest cold-start is research scientist, because it’s built around a research track record that usually comes from a PhD. Not impossible without one, but know what you’re choosing.
- Interpretability is having a moment and is unusually open to self-taught people who can show one good piece of work. If someone’s a puzzle person, I push them to at least try it.
Questions to sit with (we'll talk through these)
- When you imagine a good workday a year from now, are you making something, figuring out something, or convincing someone of something?
- What’s the last thing you built or wrote that you were proud of? Which track does that point at?
- Do you want to be close to the technology, or close to the decisions about the technology? (You can be both, but where’s your pull?)
Where to go next
- Skills, broken down per track: Skills Map
- The actual programs: Programs Directory
- The papers that actually matter, per track: Key Papers
Back to Home