Podcast Conversations between AI Local LLM Model is a unique project where several models have a conversation with each other.
This conversation can be on a specific topic, maybe even a task or research or nothing, where they simply discuss with each other without my involvement.
What do I use?
I use ollama and the available LLM models that can run with ollama. I made a python script [actually I have several versions] that allows 2 or more LLM models to have an autonomous conversation without my involvement.
I also made a reasoning script to increase their understanding abilities.
In 24 hours of running I can generate up to 11 million words, an impressive amount of text.
You can find and download all the conversations from the Resources section.
I make the podcast using NotebookLM because it is simpler and easier plus in general it makes a very good summary of large texts and brings me perspectives, ideas that I did not know or see before.
You will generally find 1-2 Podcast versions for each conversation.
Why is this?
Because the first time I usually do a general podcast and if it's ok, then I let them focus only on the discussion between the LLM models.
If the topic is not taboo then it comes out well from the first try but if it is taboo then many trial versions are needed because NoteBookLM hallucinates a lot.
How does the system work?
1. the environment is created which is very important from what I have seen. Depending on the topic I create an adequate environment for it
2. the topic is set
3. the LLM models are chosen for this conversation. Now for example there are 16 LLM models in the conversation
4. sometimes I need to make custom models for certain discussion topics.
In a conversation to imitate a discussion between real people, each LLM model can decide whether to participate in the discussion or not. It can be the entire conversation or just some things.
Very rarely a model does not participate in a conversation at all.
This conversation can be on a specific topic, maybe even a task or research or nothing, where they simply discuss with each other without my involvement.
What do I use?
I use ollama and the available LLM models that can run with ollama. I made a python script [actually I have several versions] that allows 2 or more LLM models to have an autonomous conversation without my involvement.
I also made a reasoning script to increase their understanding abilities.
In 24 hours of running I can generate up to 11 million words, an impressive amount of text.
You can find and download all the conversations from the Resources section.
I make the podcast using NotebookLM because it is simpler and easier plus in general it makes a very good summary of large texts and brings me perspectives, ideas that I did not know or see before.
You will generally find 1-2 Podcast versions for each conversation.
Why is this?
Because the first time I usually do a general podcast and if it's ok, then I let them focus only on the discussion between the LLM models.
If the topic is not taboo then it comes out well from the first try but if it is taboo then many trial versions are needed because NoteBookLM hallucinates a lot.
How does the system work?
1. the environment is created which is very important from what I have seen. Depending on the topic I create an adequate environment for it
2. the topic is set
3. the LLM models are chosen for this conversation. Now for example there are 16 LLM models in the conversation
4. sometimes I need to make custom models for certain discussion topics.
In a conversation to imitate a discussion between real people, each LLM model can decide whether to participate in the discussion or not. It can be the entire conversation or just some things.
Very rarely a model does not participate in a conversation at all.