Anthropic's RL engineering posting now asks for experimental judgment
Between the September 24 and October 3 captures, the same RL Engineering posting made controlled experiments and research contribution explicit. It also moved experience training large ML models or building their supporting systems into minimum qualifications. The earlier posting emphasized researcher productivity, performance and reliability; large-model training experience appeared as an additional strength. Both versions already supported production and research runs. The change is in advertised expectations, not proof of a changed interview process.
The constraint
Is this an infrastructure role supporting researchers, or a role expected to reason about training itself?
Another explanation
This may clarify expectations that already existed rather than change the team. It does not show that engineers have delivered a research improvement.
Evidence
- Anthropic · Research Engineer, RL Engineering
Ability to design controlled experiments and reach conclusions you and others can trust
Context
- Earlier retained Research Engineer, RL Engineering posting
The earlier role already supported production and research training, performance and reliability. Its applicant section invited engineers who wanted to learn more about ML research. Capture comparison: 2026-09-24 to 2026-10-03. These are observation bounds, not an exact edit date; the URL now serves the current posting. This comparison uses all retained text; the older capture has no explicit full-description completeness certification.
What would change this reading?
A hiring-team explanation or interview rubric showing whether controlled-experiment design now changes selection for this role.
For candidates
Show how you isolated a training problem, designed a controlled comparison and reached a reliable conclusion, alongside systems performance work.