karthink
commented
8 months ago You can already do some of this with gptel-set-topic, which limits context by Org heading. You can write some glue code that creates an Org link from selected text to a new heading with that heading.
zikajk
Hello,
I frequently engage in extended discussions with ChatGPT, often diving deep into specific details. However, I've found that these in-depth conversations can become lengthy and hard to navigate. And creating a new conversation thread will lose context so I don't do that.
Here's an idea I've been considering: Would it be possible to use Emacs, in conjunction with your package, to create a more organized interaction with ChatGPT? My thought is to have a main conversation buffer and then branch out to topic-specific buffers when seeking detailed information. This would allow for a modular discussion, making it easier to navigate between the main topic and its subtopics.
For instance, if I ask ChatGPT about the "10 largest lakes in Europe", it would provide a list. I could then mark "Lake Ladoga", use a function like
(give-me-details), and a new buffer focused on "Lake Ladoga" would open where I can continue with the conversation on this topic. Within this buffer, the conversation about this specific lake would ensue. Ideally, there should be some built-in logic: upon using main buffer again, a prompt like "thanks for the details, let's return to the discussion on the 10 largest lakes" would be sent to ChatGPT, and vice versa. Visually, this could resemble a ROAM-style map, with "10-largest-lake" as the main node and "Lake Ladoga Topic" as a connected sub-node.Do you think this approach is feasible with the gptel package? I believe it could enhance the user experience when having detailed conversations with ChatGPT.
Edit:
(give-me-details)might offer a list of choices as customizable prompts.