Remote Image Reasoning Generalist - AI Trainer ($35-$35 per hour)

## About this opportunity Mercor is working with a leading AI lab to build a large collection of visual tasks to help train advanced AI models. This project focuses on **counting objects in images**. While counting sounds easy, AI models often make mistakes. They may guess instead of counting, rely on text in the image instead of what they see, count duplicates incorrectly, or miss objects that look similar. Your job is to source images that can cause these mistakes and clearly document what went wrong. ## What you will do - Find an image that may cause an AI model to count incorrectly. The image can be original, public domain, or AI-generated - Identify what might confuse the model, such as misleading text, similar-looking objects, duplicates, or an unclear unit of measurement - Test the image with AI models using a simple prompt: **"How many X are there? Return an integer."** - Write additional prompts that give the models more context, and then label the task ## What we're looking for - **Strong written English**: Prompts need to be clear and easy to understand. - **Good attention to detail**: You should be able to notice small details in images and understand what might confuse an AI model. - **Consistency and reliability**: You will work on many tasks, so being careful and consistent is important. Experience with photography, image editing, data labeling, or other detail-focused work is a plus. ### Details - **Remote:** Work from anywhere - **Pay:** $35/hour