Getting a weird response from a looping Ask AI node that is in a subflow of a larger automation. I’m getting content completely unrelated to the inputs provided (I think). Video provided below.
This is a super weird case of the model hallucinating – there’s nothing wrong with the setup of the flow. How often is this happening? One way to identify these rare hallucinations is that the AI didn’t go off the rails in any other node or loop iteration, plus it gave the accurate output for loop iteration #6 but randomly added details below that.
If you disable cache on the ‘Ask AI’ node right below the Rewrite recommendations AI node under Show More Options (so the AI doesn’t repeat the same response given the same inputs), and resume the flow, you’ll notice it doesn’t hallucinate and gives the correct response.
As for the deleted subflow point from the video – if you delete a subflow but not the subflow node, and it’s still being used in the flow, it gets preserved to avoid breaking the flow. That’s why the Open Link brings up the same workbook even though the tab for that subflow is gone. You can keep it as is if needed, or replace it with the newer subflow from the node library. Either way, this isn’t related to the issue above.
If this keeps happening, let me know and I’ll take another look.
Every run is a version! If you want to go back to a specific version you can overwrite the current version my going to that run from the https://www.gumloop.com/history page and clicking save!
I’d recommend duplicating your current flow just in case.
Yeah see thats what i thought! But when i tried it didn’t work. I probably made a mistake though, i’ll double check and let you know if there are any issues.