Latest LLMs powering Q model
The next evolution of Q in QuickSight Q&A is here. With the release of our V3 model, advanced large language models (LLMs) are now seamlessly integrated, delivering improved accuracy and addressing many previously unanswerable questions. Users can ask questions in more natural language, without needing to rephrase them in a structured or SQL-like format.
Available now across all genBI regions (see supported regions). All existing and new topics are automatically upgraded to V3, so you can start exploring the enhanced experience today. If you are not yet a Q in QuickSight user, you can get started with a free trial here.
Better language understanding
V3 Q&A is designed to interpret naturally phrased questions with greater precision. For example, it can understand a query like "list of seniors with an almost perfect score in Data Visualization" and map the phrase “almost perfect score” to a range, such as “average Test Score between 90 and 100.” Previously, answering such a question often required creating a calculated field or explicitly specifying the score range in the query. With V3, these steps are no longer necessary, allowing for a more seamless and intuitive question-and-answer experience.
Less reliance on topic metadata
With improved language understanding, V3 Q&A reduces the effort required by authors to fine-tune topics. For example, in the past, an author might have needed to include a calculated field to convert “Duration Of Adverse Event Hrs” from hours into minutes for readers to ask a question like “how many patients had adverse event last for over 120 minutes”? Now, Q V3 can automatically interpret the semantic information to map 120 minutes to “Duration of Adverse Event Hrs is more than 2.”
Q uses information like the field name “Duration Of Adverse Event Hrs” and the Semantic Type and Sub-Type to do an on-the-fly conversion from minutes.
Real-world knowledge
Q understands concepts that don’t need to be explicitly defined in your data. While the underlying data values must exist, categorization is not always necessary. For example, if you ask Q, “ What is the market share for recurring Chase transactions since last Black Friday?”, Q automatically maps “Black Friday” to “November 24, 2023”, and recognizes that “market share” refers to the “percent of total transaction amount” in the dataset.
In another case, Q can interpret the term “failing” in the question “How many students are failing Computer Science?” as mapping to an F grade.
Similarly, with generational terms, such as “Millennials who did not complete their drug trial”, Q understands that “Millennials” refers to those aged between 23 and 38, as long as the required data, like age or date of birth, is available.
As a final example, Q can understand this question that asks about a known location or event: how many healthcare prospects do we have in the 2028 Olympic city?
Feedback?
If you encounter any issues, please log a support ticket. To help us resolve the issue more quickly, please include the following:
- A screenshot of response
- A brief description of the issue
- The request ID (available from the top-right Share menu from the Q&A pane)
Feel free to share your experience and any helpful tips in the comments, thank you!