I Gave an OpenAI Dot a Hiring Task, Then Tried to Trip It Up
added
OpenAI Dots hands-on: the creator gave an actual Dot a hiring-coordination assignment over 4 fictional intern applications (role brief, clear boundaries: organize evidence, keep missing info visible, leave decisions to the human, synthetic records only, no email or connected apps). The Dot built a tracker separating Python, testing, communication, project work, internship context, and unanswered questions — correctly keeping a claimed Python skill separate from the demonstrated project evidence (browser interface calling an existing API), and flagged that an applicant's 'three unanswered questions' were mentioned but never worded. Mid-session, the creator changed the priority to project-evidence-first (Dot reordered columns without changing any eligibility), then added 2 new applications plus an exact duplicate of applicant two — the tracker held 6 unique applicants (not 7), deduplicating the repeat, preserved the project-first organization on the new entries, and kept all source evidence attached to the right applicants.
Real handle: OpenCV University. First-hand hands-on test of a real Dot (per transcript: 'we gave an actual dot a small hiring coordination assignment'), all applicants/projects synthetic and supplied by the creator. Concrete results: (1) applicant tracker with Python/testing/communication/project-work separation; (2) refused to promote a claimed skill to demonstrated implementation; (3) adapted to a mid-session reprioritization without altering eligibility; (4) deduplicated an exact-duplicate applicant (6 unique, not 7); (5) flagged missing question wording in the supplied material. No sponsor tag. 5:22 runtime, English captions verified.
