Maybe a unique idea for a chat bot

KageSong · November 18, 2023

Okay, two disclaimers;

1 - for the sake of ensuring the respect of any personal beliefs, I believe it is best to treat this idea as a "for entertainment only" product. (If product would be the right term)

2 - I know nothing about how to begin coding anything, much less any sort of LLM. I say "GPT-based chat bot" but I have no idea if that is the right term. It just is the closest I know to what I mean. I hope this might spark interest, but I'm not sure I'd have any more input than the idea.

SO, I have read Tarot for about a decade. One of the most challenging parts of reading, is that there are a lot of relations between cards, positions, numbers, elements, genders, etc., that can give much deeper insight into any reading, but it's things that take a life time to learn. I'm interested in the idea of an LLM that is trained specifically on various tarot decks, spreads, and various tools for interpretation like what I mentioned.

I think on the surface the idea sounds... it's hard to say quite neutrally, but avoiding a skill that should be respected if you do happen to be on the more invested side of the fence. However, I struggle with learning and have found things like Bing chat to help me get my mind around things. I'm even understanding auto hot key, which videos and such haven't helped me with, because I've tasked Bing with making a script, worked through testing and debugging with Bing's help, and got to exactly the function I want. In the process, I understand SOME of the code for ahk.

So, what I would see in such an LLM is maybe a bit of a shortcut, but even in that regard, I think a personal touch is important if you are reading for more than entertainment, and it matters less if your goal is entertainment. So, in that way, it's more for me a way to help gather thoughts and information in a way that is catered and easier to consume. It would also serve for me, and I'm sure many, as a tool for learning these more advanced patterns that can take so long to learn. A stepping stone, just a really big one. I think this could easily apply over to other tools such as I-Ching, runes, Lenormand, etc.

I do not want to get into any discussion of the legitimacy of tarot or any divination. This isn't the place. I am interested in the tech discussion of such a specifically trained LLM, as I've had great results with Bing, but there's limitations due to it's breadth, and I suppose many factors, that make it difficult to really focus on the subject, and it begins to repeat, etc. I'm curious of the viability of a focused model that would be less likely to hallucinate or give statements that simply repeat an earlier statement for what it "thinks" is some narrative purpose.

Mods - I wasn't sure if this was General or Off Topic really. I didn't see a place I felt more specifically appropriate than one of the two. Sorry if it needs to be moved thanks

Skiiwee29 · November 18, 2023

<-- moved to programs, apps, and websites. -->

Biohazard777 · November 18, 2023

16 minutes ago, KageSong said:

I do not want to get into any discussion of the legitimacy of tarot or any divination. This isn't the place. I am interested in the tech discussion of such a specifically trained LLM

But, right before that you say:

15 minutes ago, KageSong said:

I think a personal touch is important if you are reading for more than entertainment

Anyhow.
It is like asking: "How can I train an LLM which would help me design a perpetual motion machine?"
You can't write code or train an LLM and expect it to complete (or help you with completing) physically impossible tasks.

KageSong · November 19, 2023

On 11/17/2023 at 9:15 PM, Biohazard777 said:

But, right before that you say:

Anyhow.
It is like asking: "How can I train an LLM which would help me design a perpetual motion machine?"
You can't write code or train an LLM and expect it to complete (or help you with completing) physically impossible tasks.

Well, you've contributed nothing, and more than that, whatever you're trying to clearly be snide about is completely lost on me. I'm certainly not trying to build a machine.

Biohazard777 · November 19, 2023

2 hours ago, KageSong said:

Well, you've contributed nothing, and more than that, whatever you're trying to clearly be snide about is completely lost on me.

Sometimes the answer you get isn't the answer you seek.
And I'm pretty sure it wasn't lost on you.

2 hours ago, KageSong said:

I'm certainly not trying to build a machine.

Then what is the point of this thread? You said:

On 11/18/2023 at 3:46 AM, KageSong said:

I'm curious of the viability of a focused model that would be less likely to hallucinate or give statements that simply repeat an earlier statement for what it "thinks" is some narrative purpose.

The simple answer is: no it isn't viable. If that wasn't clear enough from my previous reply.
When you can't tell the difference between a human hallucinating and an LLM doing it, you can't fix/tune the LLM.
Now don't get offended over "human hallucinating", the practice is subjective... relying on personal perspectives and beliefs rather than objective, empirically verifiable insights.