Become an AI Expertv2.1
0 of 0 steps Continue
On-ramp from zero (Path A)Novice track, module A8; optional, the slower version of Modules 12, 13 and 17

From employed to expert

Track: Novice (Path A, the on-ramp from zero) Time: about 15 hours a year of planning and re-running; the work itself is 3 to 5+ years. Goal: Be someone whose judgment other engineers defer to in your track. Optional: the professional path covers this in Module 17. Principles: Principle 19 (models change under you). Lab 13 is the model-upgrade drill.

The markers (from people who got there)

  1. You own a problem end to end in your track, and your manager can state your impact in one sentence. The FAANG MLE says visibility, more than raw skill, is how promotions actually happen.
  2. People outside your company know your work. A widely used repo, a reference write-up, a course module, a conference talk, or a paper. The verified experts in the research all have this: Chris Olah's explainer blog got him the Google Brain internship with no degree; Jeremy Howard's Kaggle results and fast.ai; Aleksa Gordić's YouTube channel and repos got him the DeepMind referral.
  3. You can go arbitrarily deep in one area under questioning. A staff MLE running an AMA rejected candidates who could not. Pick the question you most fear in your track and be able to answer it for an hour.
  4. You learn on a cadence. Each new model release: re-run your evals and update your mental model. Each year: one long study (a new architecture, a new RL recipe, new hardware).
  5. You are generous. The most upvoted self-taught story in the oldest thread credits "being generous helping others" for every opportunity he got. Answer questions in the communities you learned from.

A yearly loop

  • Quarter 1: pick one long study for the year. Replicate something, write it up.
  • Quarter 2: ship or contribute. One open-source PR to a project in your track, or one internal system you fully own.
  • Quarter 3: teach. A talk, a tutorial, a course module.
  • Quarter 4: re-run everything. Evals, benchmarks, your own beliefs. Update your write-ups.

Build

Project A8: the yearly plan. One page in the Module 17 format: the four markers with your current evidence for each, the biggest gap, the one artifact that closes it, the long study for next year, your three sources, and the date of your next model-release eval run. Done when every line names something a stranger could check.

The honest counterpoints

  • A master's is close to the minimum for many ML roles and the market is flooded (2025 thread). Even the self-taught staff MLE says he still feels prejudice for not having one. If you reach this stage and a part-time master's is affordable, it removes a filter. It is not required to do the work.
  • Survivorship bias is real. Of the famous "self-taught" names, only Chris Olah, Alec Radford and George Hotz truly lack a degree. Noam Shazeer and Neel Nanda are elite-school math prodigies. The repeatable path is the adjacent-job-then-sideways one, not the dropout-to-Anthropic one.

Sources

Follow-ups

When you finish this module, paste one of these into your assistant. In Claude Code write /aiexpert-tutor before it; in Codex write $aiexpert-tutor.

  1. Ask me for the evidence behind each of the four markers.
  2. Review my yearly plan and check that it names a specific artifact and date.
  3. Walk through my model-release checklist with me on a real system.

Free to use: code MIT, writing CC BY 4.0, third-party media under its owners’ terms. Licence and terms.