Agent-based approaches coupled with large language models (LLMs) are quickly transforming how we interact with databases and data warehouses. Combined, these technologies enable natural language queries to data in your application or business, eliminating the need for SQL expertise to interact with data and even facilitating seamless queries across diverse systems. (View Highlight)
Agent-based approaches coupled with large language models (LLMs) are quickly transforming how we interact with databases and data warehouses. Combined, these technologies enable natural language queries to data in your application or business, eliminating the need for SQL expertise to interact with data and even facilitating seamless queries across diverse systems. (View Highlight)
Before diving into the example, let’s talk about synthetic data. With Gretel’s models, you can make an artificial but statistically similar version of your sensitive data. This synthetic data is safe to use, thanks to math-backed privacy features like differential privacy. In our example, we’ll use both real and synthetic data to show why this privacy is crucial when letting language models access sensitive info. (View Highlight)
Before diving into the example, let’s talk about synthetic data. With Gretel’s models, you can make an artificial but statistically similar version of your sensitive data. This synthetic data is safe to use, thanks to math-backed privacy features like differential privacy. In our example, we’ll use both real and synthetic data to show why this privacy is crucial when letting language models access sensitive info. (View Highlight)
For this example, I used the Gretel Tabular DP model (notebook, docs) with an epsilon value of 5 for strong privacy guarantees that are great for regulated environments. For maximum accuracy while still maintaining privacy, you can also try the Gretel ACTGAN model (docs), which excels at working with highly dimensional tabular data to enable machine learning and analytics use cases. (View Highlight)
For this example, I used the Gretel Tabular DP model (notebook, docs) with an epsilon value of 5 for strong privacy guarantees that are great for regulated environments. For maximum accuracy while still maintaining privacy, you can also try the Gretel ACTGAN model (docs), which excels at working with highly dimensional tabular data to enable machine learning and analytics use cases. (View Highlight)
For this example, I used the Gretel Tabular DP model (notebook, docs) with an epsilon value of 5 for strong privacy guarantees that are great for regulated environments. For maximum accuracy while still maintaining privacy, you can also try the Gretel ACTGAN model (docs), which excels at working with highly dimensional tabular data to enable machine learning and analytics use cases. (View Highlight)
Here, we illustrate a “re-identification attack” where vulnerabilities in even de-identified datasets can allow an attacker to re-identify individuals by combining known attributes. Such risks emphasize the danger of sharing data stripped of direct identifiers yet containing attributes that, when combined, can lead to identification — such as the combination of an attacker who knew someone’s age, gender, and department in the example below. (View Highlight)
