metaphilabs/

remotion-video-preferences

Pairwise human preference labels for AI-generated educational video tutorials.

Source: Expert annotations from creators.metaphi.ai Video Arena Format: Each row is one A/B comparison per chapter with winner label and reasoning Agents: Three harnesses (Claude Code, Codex, Gemini CLI) compared pairwise Use case: RLHF reward model training, preference-based optimization

	Schema




	

Column Description

id Unique vote identifier

task_id… See the full description on the dataset page: https://huggingface.co/datasets/metaphilabs/remotion-video-preferences.