Task Writer: Project Agentic-MME

About the project

Agentic-MME tests whether AI models can investigate an image rather than just look  solve a task, the model has to crop, rotate or enhance the image so a small detail becomes readable, then search the web for a fact, and then check each against the other before answering.

• Use high-resolution images where the key detail is small or degraded to test frontier AI models on the raw image and find what they miss, then build tasks around those gaps

• Write realistic prompts that need image work and at least one web search

• Record the shortest solution path step by step, with the image operations used and a credible source for each search

• Give one exact golden answer, a clear answer format and any accepted variants

• Strong attention to detail

• Excellent writing skills

• Skill at web research, with a habit of using primary, authoritative sources

• Advanced degree holders

• Experience working on rubrics is a plus but not a must-have