Good question indeed. I haven’t stress tested this action. When asking “list all records tagged with xxxx”, it only responds with some of them, not all of them. The responses are accurate, but not complete.
Thoughts? @Jeremy
1 Like
Regarding the “Answer question about the table”:
It might help to understand the mechanism - the action pulls in one row of data most relevant to the question given and use that as context to answer the question. For example, you have a list of FAQ with the column question and answer - and present it as a front desk helper.
Currently not set up to do rollup/aggregation over many rows like Roberts use case.
It tends to do well when the columns contain more text like descriptions than categorical data like tags or numbers.
More feedback about this is welcome in OpenAI Integration: Ask a Question about a Table action
Regarding Robert’s use case, we have some internal experiments about prompt based, dynamic rollups/aggregation we’re playing with but nothing for public use right now.
Would love to hear use cases and expectations around this kind of tech.
4 Likes