January 2024 Summaries

Gretel has introduced Transform v2, a model that enables 10x faster speed and supports arbitrary PII entities. It also offers completely flexible transformation rules. Users can leverage this new version to de-identify HIPAA datasets in just a few clicks. The beta release of Transform v2 allows users to detect custom entity types specific to their business and define custom transformation rules for generating data that meets the exact format and logic required. This makes Transform v2 adaptable beyond de-identification, covering most dataset pre and post processing use cases. Additionally, Transform v2 is faster than its predecessor (100MB/minute in internal load tests) and integrates fully with the Gretel ecosystem including synthetic data models, connectors, and workflows.

Gretel is a platform that enables organizations to generate synthetic text data with differential privacy guarantees, making sensitive natural language data available for various use cases while maintaining individual privacy. The platform uses state-of-the-art large language models (LLMs) and applies differential privacy during model fine-tuning through the DP-SGD algorithm. Gretel's synthetic text generation approach is particularly impactful in industries like finance, healthcare, and customer support, where handling sensitive data securely is crucial. The platform provides a reliable measure of data utility without compromising private data using its SQS (Synthetic Quality Score) evaluation metric.