Please share your new Idea for Generative AI using LLM.
I am grateful for the opportunity to share this idea with this brilliant and altruistic AI community. It was not easy for me to overcome my fear and self-doubt. Thank you for your attention and I welcome any feedback or assistance.
As a novice AI Engineer, my goal is to use my diverse and extensive IT Engineering skills to build a responsible Generative AI LLM system (Just LLM) that minimizes systemic and historical biases in America.
Historical data often reflects the bias in favor of the dominant or majority group in American society. For example, if we build an HR system powered by GenAI, it might pick Caucasian males for promotion more frequently in the workforce, since the pretrained data mirrors this history (when women and minorities joined the workforce much later and in lower numbers).
To solve this, there are two possible ways:
When we train the LLM, we need to create datasets that have equal representation of all groups and intersections.
Alternatively, we could remove gender, age, race, etc. from the system and treat them as PII with an identifier that is not used in the training data (although this might not be acceptable to some stakeholders).
Either way, I think this would be a very useful feature for a pretrained LLM model. I have over 25 years of IT engineering experience, but I have been learning GenAI and ML for only a few months. I am not sure about the best approach or the steps involved to solve this. Is there any community support for such a project? I think RAG could work along with Chain of Thought or some other methods. How can I build a prototype? What are the costs for the platform and tools for something like this? I would appreciate any help or guidance from anyone with expertise or interest in this area.
“Content crafted with the aid of AI”