Sponsored
$5/week ↗
Work & career

I Gave an OpenAI Dot a Hiring Task, Then Tried to Trip It Up

@OpenCVUniversity

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.

See all 14 →
Fusion Events' Dot audited his backlinks and found a broken one

Fusion Events' Dot audited his backlinks and found a broken one

A pleasant conversation isn't my benchmark for an AI assistant. Can it do something useful with the work I've already got going? In this demo, I called my Dot and gave it a real task: audit the backlinks to Fusion Events, identify worthwhile sites to approach, and work out how we'd approach them. The useful bit was wonderfully unglamorous. A broken backlink. An audit I could review. A correction email drafted with my signature and left unsent for me to approve. Nobody needs another robot confidently announcing that it has revolutionized their business. Fixing a broken link will do nicely, thanks. What interests me here is the context. I've already spent time working through projects in ChatGPT. Being able to pick up that work by phone matters more to me than starting another assistant from zero and explaining my business all over again. Thats how Im judging these tools: give them a specific job, inspect what comes back, and keep control of anything that goes out in my name. Want to try the same task? Give your assistant your exact website address and ask it to verify existing backlinks, separate confirmed findings from gaps in the data, and draft one specific correction request. Check the source page yourself before sending anything. If it cant show you the evidence, it isnt a verified finding. Start with a loose end in your business. See whether the assistant can actually help you close it. #dot #openai #coding #ai #aitips