What types of data points do you use (such as labels or other characteristics)?

We use multiple types of data points depending on the capability:

• Unlabeled text data drawn from a variety of domains and languages
• Text data paired with labels obtained by three methods: (1) annotated by internal specialists, (2) derived from user interaction data in accordance with our privacy policy and data governance controls, and (3) generated synthetically by AI models
  • Labels are defined based on ideal behavior of product features (e.g., grammatical error correction, paraphrasing, etc.)
• Image data, including both real images and AI-generated images