Summary of Safesora: Towards Safety Alignment Of Text2video Generation Via a Human Preference Dataset, by Josef Dai et al.
SafeSora: Towards Safety Alignment of Text2Video Generation via a Human Preference Dataset
by Josef Dai, Tianle Chen, Xuyao Wang, Ziran Yang, Taiye Chen, Jiaming Ji, Yaodong Yang
First submitted to arxiv on: 20 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Databases (cs.DB)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel dataset, SafeSora, is introduced to address the risk of harmful outputs from large vision models (LVMs) in text-to-video generation tasks. The dataset captures human preferences along two primary dimensions: helpfulness and harmlessness, with sub-dimensions and categories providing structured reasoning for crowdworkers. The dataset consists of 14,711 unique prompts, 57,333 unique videos generated by 4 LVMs, and 51,691 pairs of preference annotations labeled by humans. This allows for training text-video moderation models, aligning LVMs with human preferences, and developing alignment algorithms. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A team of researchers created a new way to help computers generate videos that are safe and helpful. They made a special dataset called SafeSora that has lots of examples of what makes a video good or bad. This helps computers learn what is right and wrong. The dataset includes many different prompts (ideas) for making videos, the actual videos themselves, and how people think they should be rated. This can help create better computer programs that make helpful and safe videos. |
Keywords
» Artificial intelligence » Alignment
