Overview
Mercor partners with leading AI labs to train and improve frontier models. In this role, you'll create challenging image-based counting tasks—original outdoor photographs paired with reasoning-based prompts—designed to expose weaknesses in AI vision models and help improve their accuracy.
What You’ll Do
- Take original, unedited photographs of busy real-world outdoor scenes (streets, parking lots, buildings, parks, markets)
- Write 3–5 counting prompts per image that require reasoning (conditional filters, spatial relationships, comparing similar objects, zero-answer prompts)
- Verify that each prompt has one correct answer that 9 out of 10 people would agree on immediately
- Outline and mask every counted object in an annotation tool
- Check answers object by object and submit accurate counts
- Review and act on feedback from project reviewers
- Flag unclear guidelines rather than making assumptions
Requirements
- Sharp eye for detail and patience to accurately count dozens of objects
- Own phone or camera with access to busy real-world outdoor scenes
- Clear, precise writing—prompts must avoid vague terms like 'closest' or ambiguous color descriptions
- Comfort with careful, repetitive checking and detailed verification work
- Ability to take original, unedited photographs (no stock, social media, AI-generated, or edited images)
- Willingness to engage with reviewer feedback and iterate
Who Should Apply
This role suits people with a meticulous eye, strong visual attention, and genuine access to diverse busy outdoor scenes. It's ideal if you're detail-oriented, enjoy writing precisely, and have the discipline to do repetitive verification work with no margin for error. It will frustrate you if you prefer broad strategy over execution, dislike photographing real-world scenes, or struggle with ambiguity—the project requires you to flag unclear guidelines rather than guess.
Salary Insight
Pay is per-task based on object count: $60 for 0–24 objects, $70 for 25–49, $80 for 50–79, $100 for 80+. No stated hourly rate; earnings depend on your pace and task complexity. Since tasks go through review and rejected work reduces pay, careful first submissions matter significantly. Research similar annotation/labeling projects to understand typical completion time and whether the per-task rate aligns with your local wage expectations.
About Mercor
A marketplace placing specialists into paid remote contract work with AI companies. One profile and one AI interview, reused across every role you apply to.
Remote policy: Fully remote, worldwide, on hourly contractor agreements rather than employment. But payment eligibility is the thing to check before you invest any time, because it is decided by country and there is no way around it. Mercor pays through Stripe Connect in…
How hiring works at Mercor →Required Skills
Compensation
$20–$80 per task
- Location
- Remote
- Engagement
- Contract
- Posted
- 6d ago
Opens Mercor’s listing on MyRemoteJobs — we don’t collect applications ourselves.
About Mercor
A marketplace placing specialists into paid remote contract work with AI companies. One profile and one AI interview, reused across every role you apply to.
Remote policy: Fully remote, worldwide, on hourly contractor agreements rather than employment. But payment eligibility is the thing to check before you invest any time, because it is decided by country and there is no way around it. Mercor pays through Stripe Connect in…
How hiring works at Mercor →Will your CV get past the filter?
Most applications are rejected by software before a person ever reads them. Scan your CV using our ATS Checker tool, see exactly what's blocking it, then fix it in the CV Builder and apply knowing it'll get through.
Scan my CVGet roles like this by email
Tell us what you do and we'll email you when a similar remote job is posted. No account needed.
See our jobs more often in your search
Add MyRemoteJobs as a Preferred Source on Google. Free, takes seconds, and you stay on this page.
Application Tip
Your photos are the primary filter—submit one strong outdoor scene with genuinely challenging prompts that a current AI vision model would likely fail. Don't pad the task with easy or ambiguous prompts. After you pass the assessment and onboarding quiz, your first qualifying task signals everything the team needs to know: whether you can find busy, countable scenes and write precise reasoning-based prompts that actually expose model weaknesses. Lead with originality and accuracy, not volume.
Sourced from MyRemoteJobs · verified against the original posting
Similar open positions
English Language Expert (United States)
micro1
VerifiedAI Conversation Evaluator
micro1
VerifiedEmergency Medicine Physician
micro1
Verified