I know I can chat about documents with AI services in DT4 now, which is great. So far I have only done this by selecting a singular document in my database.
I was wondering if it’s possible to chat with AI in an „expert mode“. What I mean is, I could define a few documents from my database as background information for an upcoming discussion. I imagine that discussing a specific document in my database with AI would be much more fruitful if the AI considered my predefined documents as background information.
Is there a proper way to accomplish this in DT4 currently? Do I just have to select more than one document in DT4 (even if they are in different groups) and start chatting? How does DT know which document is background and which is the one I‘d like to discuss?
You have to explicitly mention the selected documents in your prompt and could reference them by their order or name. The order is usually sufficient and more efficient.
I love running role-playing games, especially those set in the Middle-Earth universe by J.R.R. Tolkien. While preparing games for my favourite system, The One Ring, I often use AI bots for brainstorming and inspiration. I then try to create a compelling plot, including riddles to solve, secrets to uncover, and interesting twists.
While this approach works ok for now, the AI bots frequently reference outdated rulebooks from the internet or irrelevant systems, lacking knowledge of my game’s history. By directing the bot to my own rulebooks and scripts of my gaming sessions, I believe the quality of its responses would improve significantly.
Thanks Christian, if I understand your comment correctly, I need to start each chat with a prompt pointing to the documents I like to be considered. Then I can jump / add a link to the document I want to discuss and start chatting. Is that about right?
I‘ve used different models from Claude, Mistral, ChatGPT and Perplexity. I can’t remember which ones exactly. Mostly the not so expensive ones, small to medium and not the newest.
Okay. I was curious as the context window matters, with a larger one being able to process and “remember” more information during the course of a “conversation”. While a commercial model still has limitations, they have much larger context windows.
PS: This was my recent addition to the Appendix in the built-in Help and manual…
Just a reference of the models DEVONthink 4 currently supports with the context window size and reported functions, e.g., vision. And yes, this list will be updated with each release.
