Send CathInfo's owner Matthew a gift from his Amazon wish list:
https://www.amazon.com/hz/wishlist/ls/25M2B8RERL1UO

« previous next »
Pages: 1 [2]

Author Topic: Traditional Catholic AI update  (Read 5186 times)

0 Members and 1 Guest are viewing this topic.

Offline TomGubbinsKimmage

Re: Traditional Catholic AI update
« Reply #5 on: November 17, 2025, 12:12:19 PM »
Why don't you like the low context window?
Logged

Offline Angelus

  • Supporter
Re: Traditional Catholic AI update
« Reply #6 on: November 17, 2025, 02:40:18 PM »
Essentially, your problem is the "consensus" model of truth that the general LLMs use. To overcome this, I use Google's NotebookLM (the consumer product) which uses a RAG model. I have uploaded all my own curated sources, including a customizable AI rules docuŠ¼ent.

It works very well at overcoming the problems with a general LLM. When I run into any problems, I adjust the rules docuŠ¼ent. However, the rules need to be very clearly stated, and you need to be careful that the rules logically harmonize, otherwise the AI might be forced into contradictory situations. These self-contradictory situations can cause the AI to become argumentative or not answer some questions. Best to start out with very few rules and add as you need them.

I suggest you try using NotebookLM as a test. See if you can get it to do what you are trying to do. If so, build up your own NotebookLM-like system on your own server.

https://thenewstack.io/how-to-deploy-an-open-source-version-of-notebooklm/
Logged

Offline St Giles

  • Supporter
Re: Traditional Catholic AI update
« Reply #7 on: November 17, 2025, 09:26:41 PM »
Dumb question: can you edit the context window size? Someone wrote that size limit into the program, right? So, can you change that?


I was afraid that even without biases, it may still take a team of people to train your model, and given the fallible and personal/biased human factor, it would be best to include the inefficient process of having everyone check each other's work to iron out each person's fingerprints from the model.


Is everything sufficient on the hardware side of things?
Logged

Offline Matthew

  • Mod
Re: Traditional Catholic AI update
« Reply #8 on: November 17, 2025, 09:57:57 PM »
Quote from: St Giles on November 17, 2025, 09:26:41 PM
Dumb question: can you edit the context window size? Someone wrote that size limit into the program, right? So, can you change that?


I was afraid that even without biases, it may still take a team of people to train your model, and given the fallible and personal/biased human factor, it would be best to include the inefficient process of having everyone check each other's work to iron out each person's fingerprints from the model.


Is everything sufficient on the hardware side of things?

No, the context window is part of the model. I'll admit, I thought the same thing at first. But there is supposedly a model with a larger context size -- I have some things to look into at this point. I'll definitely get back to you on it.
Logged

Offline Freind

Re: Traditional Catholic AI update
« Reply #9 on: November 24, 2025, 06:32:39 PM »
Nobody mentioned gab.ai here.

Why?
Logged
Pages: 1 [2]
« previous next »