JobFrame · Ladder · DSAISE.GEN.M3
Data Science / Artificial Intelligence / Software Engineering · General
The full level-by-level ladder — every rung side by side, current level (M3) highlighted.
Pay by level
Same pricing source as the profile's ticker header — family multiplier + inherited focus differential over the function×level base.
National-base pricing — this family's function differential is pending review, so dollar levels are not function-adjusted (level-over-level % is unaffected).
Pay rises with level — the structure that explains the most pay variance.
Pay figures are modeled estimates from our market equation — level-differentiated, with honest-wide bands. They refresh as new leveling evidence lands; treat per-role dollars as directional.
Mastery at this level
M3 — Foundational · does the basics reliably, by the book
Building reliable, well-specified systems
- Instruments training runs to catch agents exploiting reward loopholes
- Audits datasets for coverage of the real population before trusting outputs
- Runs stress and adversarial tests before declaring a model ready
- Tunes exploration and curricula so agents learn intended behavior
What’s next — Proficientat M3 → M4
Shifting from making a system work correctly to making it defer to and learn genuine human intent under uncertainty
What to learn
- Objective uncertainty and corrigibility theory (deference, interruptibility)
- Inverse reinforcement learning and preference-learning-from-feedback methods
- Formal fairness definitions and their trade-offs
What to practice
- Designing objectives that yield corrigible, interruptible agents
- Building human-feedback loops that update the model's goal estimate
- Encoding and testing fairness constraints across groups
- Facilitating interdisciplinary review of design decisions
From Lead An AI / ML Research Team — the closest guide to this family (semantic match; stage bands from the canonical level binding).
No JFM factory canon (responsibilities × level, level guidelines × level, skills) for this profile yet.