LATAM Software Engineers: Coding Tasks for AI Evaluation
Terac, Argentina
What you will do
- Review realistic programming tasks designed for AI evaluation
- Assess the accuracy and structure of various evaluation harnesses
- Walk us through your thought process while checking code validity
- Provide actionable feedback on how to improve task complexity and realism
What we're researching
We're running a paid study on the effectiveness of coding tasks used to evaluate AI agents. Our team is building a comprehensive suite of programming environments designed to test complex software capabilities. We want to ensure these evaluation frameworks are realistic, accurate, and properly calibrated.
How it works
You will review a series of proposed coding tasks and assess their suitability for testing AI agents. We will ask you to verify the logic of the evaluation harnesses and provide feedback on their realistic application. You will share your screen to walk through the environments and point out potential flaws or improvements. The session involves a mix of code review, technical discussion, and direct feedback on the task structures.