What is Generative AI?
Generative AI is a category of artificial intelligence that creates new content rather than only analyzing or classifying existing data. This includes generating text, software code, images, and audio from natural language instructions (prompts).
For many years, most enterprise AI was analytical. A data scientist would train a machine learning model on credit card transactions to classify which ones were fraudulent (pattern recognition). The model produced a simple output: fraud or not fraud. Generative AI works differently. Built on large foundation models (like large language models and diffusion models), it learns the statistical patterns of its training data and uses them to generate new outputs.
The Foundational Architectures
Generative AI relies on different neural network architectures depending on the type of content being generated.
1. Transformer Models (Text and Code)
Most current models for generating text and software code use the Transformer architecture (the foundation of LLMs like GPT-4). Trained on large amounts of public text and source code, these models learn the structure of natural and programming languages. A data engineer can prompt the model, “Write a PySpark script to deduplicate this customer table,” and the model will return Python code. That code still needs review and testing before it runs in production.
2. Diffusion Models (Images and Video)
Many image and video generators use diffusion models (like Stable Diffusion). During training, an image is progressively corrupted by adding random noise until it is unrecognizable. The neural network is trained to reverse that process, removing the noise step by step to recover the image.
When a user prompts the model to “Generate a picture of a futuristic data center,” the model starts from random noise and iteratively denoises it, guided by the prompt, until it produces an image that matches the request.
Generative AI in the Data Enterprise
Generative AI is changing how organizations interact with their data infrastructure.
Historically, extracting insights from a Data Lakehouse required an analyst who knew SQL. Today, lakehouse platforms (like Dremio) offer text-to-SQL features. A business user can type, “Show me the top five performing sales regions from last quarter compared to the previous year,” and the AI agent translates that sentence into a SQL query, runs it against the lakehouse, and can return a chart. Results are more reliable when the agent works from a well-defined semantic layer, and users should still be able to inspect the generated SQL.
Security and Governance
Deploying generative AI in the enterprise carries governance risks. If an employee pastes proprietary source code into a public chatbot to debug an error, that code leaves the company’s control and may be retained under the vendor’s consumer terms.
To reduce this risk, enterprises run open-weight models (like Llama 3) inside their own VPCs (Virtual Private Clouds), or use enterprise API agreements with contractual data retention and training restrictions.
Summary of Technical Value
Generative AI moves machines from classifying data to producing content. By using large foundation models to generate text, code, and media on demand, it can speed up many kinds of knowledge work. When securely integrated with the data lakehouse, it lets non-technical users query large datasets in natural language, while governance controls what data the model can see.
Learn More
To learn more about the Data Lakehouse, read The Lakehouse Built for Everyone by Alex Merced. You can find this and other books by Alex Merced at books.alexmerced.com.