ChatGPT announced a beta feature that lets companies include their own data in its Deep Research results. Users can point to files stored in GitHub, Microsoft OneDrive and Sharepoint, and Dropbox. Note that including data in query results isn’t the same as using that data to train a model.
Microsoft is offering a low-code tool that lets non-technical users train AI models on their company’s own data. Microsoft 365 Copilot Tuning will be in limited release this June. Other announcements at the company’s Build 2025 event include multi-agent orchestration and new agent-building tools for developers.
Since we’re all about data access today, let’s also report that analytics platform Sigma is adding a File Column Type that can store unstructured files such as PDFs and images in Sigma tables, bringing structured and unstructured data together. I knew you’d be excited.
Composable CDP Hightouch is introducing same-session personalization, providing the ability to combine warehouse data and real-time signals during real-time interactions. It requires caching audience data outside of the data warehouse, bending the company’s “no copy” rule. This blog post explains how they do it.