You want to use a second window with a second chat to avoid placing Intermitt Questions in an ongoing Chat to avoid throwing of the LLM from the core topic.
An LLM will always fully incorporate within a running Chat all questions. The more off topic or off the core line of the chat an arising question is, the more the LLM will alter away from the "red line", the core topic, of the chat.
Those are clarifying questions such as about names, dates or word definitions and synonyms. In the same way that the tiny little word "but" alters the importance of the entire statement before to potentially even irrelevance a LLM chat can be thrown off by a far bit by everything we'd start in a human interaction with "By the way..."
I assume that this effect will increase having a set of Expert Systems, as the Cyberdeck reassembles from even the stand alone version on, ready to be build.
So, you have created the Obsidian-Deck and RAGed a to your hardware fitting LLM onto a set of relative texts such as the works of Socrates and European Romanticism all from OpenLibrary. The Vault of the Obsidian-Deck is filled with the works of an online Cyberpunk role playing game forum.
You now can cluster the forum content into Short Stories, Adventures, Rule Criticism, Role Play Scripts using the Obsidian-Deck and use the RAGed LLM to analyze the content giving you the perspective from European Antique philosophy and a specific genre in poetry and art.
You also have a model that is RAGed with the Vault content. This one will answer from the forum content perspective.
The two AIs will give really, really different answers, because their data points are fundamentally different.
That makes sense, right?
Except a question that appears to be slightly off is actually, as in a good American Jury Court Movie, placed to point into a specific direction, it is better to ask a second AI and than use that answer in the other actual core chat.
#cyberpunkcoltoure #deggers
Open Source Software Research. Underground Labs...
