I’m exploring the capabilities of the new gemma3 model within DT. Using the AI Inspector’s Chat funciton It works, but highly randomly, on a simple querry of “name the selected document”, “name this document”, and other queries as outlined in AI inspector does not get the selected document - #2 by cgrunenberg . However, I continue to find it answering with the name of a previously sleected document. Yet when I shut down DT, re-select a new doc, and then ask, its right that first time.
How can I get the AI to consistently correctly answer such a basic question? (I’m doing this to learn its capabilities before building it into scripts.)
I asked the AI chat to show me the info DT had sent to it, and it provided a linkk to, and contents of, a PREVIOUSLY selected item. Is it possible that DT is not changing the info it sends to the AI when the selected item is changed?
You haven’t provided any information about the AI provider and model you’re using. And what you see as a “simple query” is only obvious to you as a human.
No worries.
Local AI models vary more in capabilities. Also, the way they’re accessed can affect this.
Also, small local models are much more prone to hallucinate.
What are you specifically trying to accomplish?
Do note we continue to make improvements and fine-tune things.
When I start in a document and follow links to another document then the selected and previewed documents are different for me.
Another situation when the reveal shortcut is useful is when I’m in a Smart Group, such as Unread, want to insert a backlink into the annotation file before using the AI chat.
I’ve just been selecting (not opening) random items in the item list and running the querry, with the inconsistant results mentioined above.
Re: bjornivarsson’s comment on CMD-R, I check the documentation but don’t really see what that does other than as an inclusion for a URL, and it does not seem to do anything when I have a slected item and hit CMD-R. How can I use that to troubleshoot this situation?
Bluefrog asked: ‘What are you specifically trying to accomplish?’ At this point I’m simply trying to understand how the AI works with DT so I can expand into useful scripts. But I’m stymied since, as I understand it, the chat SHOULD be able to consistently identify the selected document (as a starting point). This is why I’m asking what info DT actually sends to the built-in AI chat, so I get a better handle on what’s happening invisibly. Any and all suggestions appreciated!