@saimagnate ↔ @yogicoder2 conversation
Saim@saimagnate
26 views
Aug 07, 2025
~1 min read
Advertisement
Saim@saimagnate
Today I can proudly say that a 100% of scripts for all my faceless channels (except branded) are fully done with my AI systems.
Me and my team have managed to eliminate complete human dependency, saving thousands in cost.
And no, you won’t be able to tell the difference no matter how much of an AI/language expert you are.
I have built and perfected this system over a span of 6 months and have finally rolled it out across board.
Me and my team have managed to eliminate complete human dependency, saving thousands in cost.
And no, you won’t be able to tell the difference no matter how much of an AI/language expert you are.
I have built and perfected this system over a span of 6 months and have finally rolled it out across board.
Yogi coder@yogicoder2
Is this something similar to one shot learning with very very detailed prompts refined over time ?
Am right now in playground mode adding multiple examples in one shot and iterating over prompts.
The issue I see is
1. Benchmarking similarity with a source bunch of scripts ...
Am right now in playground mode adding multiple examples in one shot and iterating over prompts.
The issue I see is
1. Benchmarking similarity with a source bunch of scripts ...
Saim@saimagnate
This is the full system:
1) Custom gpt (trained on model scripts and my handcrafted scriptwriting guide) that asks a VA questions about the script
2) VA answers the questions and sends a pdf that contains deep research data (we’re currently using gemini for deep research, will give gpt 5 a shot as well) on the topic and details like desired word count, angle and sponsor info
3) Custom GPT outputs a detailed outline, guide and info dump that the VA takes over to claude (currently using opus 4) and sends it along with the deep research pdf
4) Claude outputs the final script
This was as transparent as I could get, hope it helps.
1) Custom gpt (trained on model scripts and my handcrafted scriptwriting guide) that asks a VA questions about the script
2) VA answers the questions and sends a pdf that contains deep research data (we’re currently using gemini for deep research, will give gpt 5 a shot as well) on the topic and details like desired word count, angle and sponsor info
3) Custom GPT outputs a detailed outline, guide and info dump that the VA takes over to claude (currently using opus 4) and sends it along with the deep research pdf
4) Claude outputs the final script
This was as transparent as I could get, hope it helps.
