LAION and Open Datasets Quiz

Answers and explanations appear as you take the quiz.

1. What does LAION primarily distribute in its image-text datasets?
   • URLs and metadata pointing to images on the web
   • The actual image files
   • Only text captions with no images
   • Pre-trained model weights
2. Roughly how many image-text pairs are in LAION-5B?
   • 5 million
   • 5.85 billion
   • 500 million
   • 50 billion
3. Which model did LAION use to filter image-text pairs for alignment quality?
   • GPT-3
   • Stable Diffusion
   • CLIP
   • ResNet-50
4. Which prominent open generative model was trained using LAION data?
   • BERT
   • AlphaFold
   • Whisper
   • Stable Diffusion
5. What issue led LAION to withdraw LAION-5B in December 2023?
   • Researchers found URLs to images suspected to be child sexual abuse material
   • The dataset contained no English captions
   • The CLIP model could not read any image URLs
   • Every image in the dataset had been duplicated
6. What ongoing technical weakness results from LAION storing only URLs rather than images?
   • Slower CLIP filtering
   • Link rot that degrades reproducibility over time
   • Higher storage costs for LAION
   • Inability to support multiple languages
7. Which precomputed representation can support fast similarity search across LAION-5B?
   • Full copies of every source image
   • Audio transcripts aligned to each image
   • CLIP embeddings for the image-text data
   • The complete weights of every trained text-to-image model
8. Which three language groupings does the LAION-5B paper describe?
   • English, Spanish, and French only
   • Text-only, video, and speech datasets
   • Training, validation, and test images
   • English, multilingual, and language-undetermined data