Gretel

The future of privacy-preserving generative AI is bright. AI teams can now use Gretel with #Azure AI Foundry to design and generate secure synthetic datasets tailored to their specific business needs. This integration significantly reduces costs and time compared to traditional data labeling methods, while maintaining robust privacy and compliance standards. Enterprises are already leveraging Gretel to build specialized Small Language Models (SLMs), enhance reasoning abilities in Large Language Models (LLMs), and scale data generation from limited real-world examples. “EY is leveraging the privacy-protected synthetic data to fine-tune Azure OpenAI Service models in the financial domain," said John Thompson, Global Client Technology AI Lead at EY. "Using this technology with differential privacy guarantees, we generate highly accurate synthetic datasets—within 1% of real data accuracy—that safeguard sensitive financial information and prevent PII exposure. This approach ensures model safety through privacy attack simulations and robust data quality reporting. With this integration, we can safely fine-tune models for our specific financial use cases while upholding the highest compliance and regulatory standards.” Check out the announcement for more details: https://lnkd.in/d5CMVW3y #SyntheticData #Privacy #AI

Excited to see Gretel recognized as part of the data transformation layer for AI-powered banking. Embedding privacy at the foundation of the stack isn’t just smart for banking—it's the way forward for every industry. #SyntheticData

Heading to re:Invent post-Thanksgiving. Booth 1605, the same year that Don Quixote was published.

At Gretel we are hiring for 15+ immediate open headcounts, across everything from Applied Science, Sales, and Marketing. At this early stage in AI, it is difficult to predict where AI will go exactly, but we can be certain it will require safe, high-quality, and domain specific data. So, come help us solve these data bottleneck problems and accelerate the whole AI ecosystem. Feel free to DM me directly if you find one of the roles a particularly good fit, or even if you don't see a specific role but interested in the problems we're solving. Better data makes better AI. *No recruiters please*

Gretel's SDK integration makes it easier to fine-tune OpenAI models on Microsoft Azure with synthetic data and built-in differential privacy. Whether you're starting with your own data or generating new datasets from prompts, this integration lets you safely work with domain-specific or sensitive data. Looking forward to seeing how teams use this to unlock new possibilities! ?? https://lnkd.in/g8FxYXYv

This is one of the most significant breakthroughs we have had at Gretel since the beginning of the company. ?? Whether your data is text, structured, or unstructured- you can get SOTA privacy and accuracy. Imagine safely unlocking insights from sensitive datasets like medical transcripts, call center logs, or financial data, all while ensuring mathematically-proven privacy safeguards

Today, we’re excited to announce a new measurement feature: Personally Identifiable Information (PII) Replay—the latest addition to our suite of privacy metrics that ensures synthetic data is private by design. With PII Replay, you can easily identify sensitive information in your original training data and track how often it appears in your generated synthetic data. This powerful feature empowers you to quantify privacy risks and take action to mitigate them effectively. ?? Learn more about how PII Replay ensures privacy for your data: https://bit.ly/4fFuVw3 #AI #Privacy #SyntheticData

We're excited to announce that Gretel Navigator Fine Tuning (NavFT) now has advanced differential privacy capabilities. This means you can generate high-quality synthetic tabular data—including numerical, categorical, free-text, and event-driven fields—while ensuring mathematical privacy guarantees. * Why This Matters * - Data Privacy: Differential privacy provides formal guarantees against data leakage, crucial for handling sensitive or regulated information. - Compliance: Helps meet strict data privacy regulations without compromising data utility for analysis and sharing. - Versatility: Supports various data types with models tailored to your dataset—NavFT for mixed types, Gretel GPT for free-text, and Tabular DP for numerical and categorical data. * Balancing Privacy and Utility * Our experiments show that while DP enhances privacy, the synthetic data remains valuable for analytics and machine learning tasks. Even with stronger privacy settings, models trained on differentially private synthetic data performed comparably to those trained on non-private data. * Key Tips * - Optimal Dataset Size: Larger datasets (thousands of samples) work best with DP. - Privacy Settings: Adjust epsilon values (ε ≥ 8) to balance privacy and utility, especially for complex or smaller datasets. - Batch Size: Increasing batch size can improve results while maintaining privacy, but watch for memory constraints.

?? Introducing Sample-to-Dataset: Gretel’s new workflow in Data Designer lets you generate large-scale, diverse synthetic datasets from just a few sample records. ?? Automatically expand the problem domain as a human expert would ?? Maintain your data's exact format, style, and logic ?? Experiment faster with preview-and-iterate workflows Imagine going from 5 sample records to 1,000 high-quality, structured records—perfect for fine-tuning AI models or balancing datasets. ?? Dive into the details: https://lnkd.in/eGqEvhfT #SyntheticData #AI #DataScience

Introducing Gretel Navigator Fine Tuning with Differential Privacy. Our model of choice for generating synthetic tabular datasets now features mathematically-proven safeguards for privacy — ideal for numerical, categorial, or even free-text columns. ?? ?? Discover how this breakthrough can help enterprises unlock sensitive datasets to surface powerful insights and trends without compromising utility: https://lnkd.in/d4MVmcVk #AI #Privacy #SyntheticData