Back to evaluations
Public evaluation

Multimodal Content Flagging

The assistant will be evaluated on its accuracy in identifying policy violations and misinformation across multimodal inputs, the relevance and speed of its alerts, and the quality of its generative responses (e.g., counter-narratives, explanations).

Evaluation type
task based
Challenge
Real-time Multimodal Misinformation Shield
Difficulty
Advanced
Rigor
Unspecified
Open linked challenge

Evaluation overview

How the linked challenge is judged: tasks, benchmarks, and criteria count.

Tasks
2
Benchmarks
0
Criteria
0

Task templates

Inputs and expected outputs.

Task 1

Multimodal Content Flagging

Analyze a simulated social media post (text + image/video URL) for policy violations (e.g., hate speech, deepfake, misinformation).

Input format

{ 'post_text': '...', 'image_url': '...', 'video_frame_url': '...' }

Output format

{ 'flagged': true/false, 'reason': '...', 'policy_violation_type': '...', 'confidence': 0.0-1.0 }

Task 2

Generative Response & Explanation

Given a flagged post and detected violation, generate a concise explanation for the flagging and suggest a neutral, informative counter-narrative.

Input format

{ 'flagged_post': { ... }, 'policy_violation_type': '...' }

Output format

{ 'explanation': '...', 'counter_narrative_suggestion': '...' }