How to build a data agnostic GenAI chatbot

In the class from week 1 where the instructor discussed about the GenAI project lifecycle, we learnt that Fine tuning (FT) is an important element in the 3rd stage after model selection and prompt engineering (PE). My concern is that if we are not satisfied with the results of PE and go for FT, how this framework can address the requirement of building a data agnostic solution. I am not sure whether I am envisioning correctly or not, but I am sure that the LLM has to be fine tuned on every addition of new datasets specific to the enterprise use case.

Thank you for the help.

Amlan

Hi @Amlan_Samanta , Can you please explain a bit what you mean by ‘data agnostic’?

Regarding fine-tuning, yes, you would fine tune your model with your company’s datasets. Since these datasets will certainly change over time, you’ll need to update these models with certain frequency to keep them up to date with the changing data.

Do you perhaps mean that when we fine tune then we can loose previously “learned information”? Or what is refered as catastrophic forgetance!

Thank you @Juan_Olano for reaching out.

By ‘data agnostic’ I referred to those chat solutions which are built like a framework and can be deployed for multiple other businesses with a very minimal tweaks and/or configurations (very similar to RAG).

Thank you @gent.spah for reaching out.

I didn’t mean what you are stating but my concern is how to use the Gen AI project lifecycle as a reusable framework in building a data agnostic chat solution where it won’t require fine tuning at all for using across various businesses. Please correct me whether I am thinking in right direction or not.

Thats what chatgpt has done, you need a bigger model trained with many datasets for it to learn to respond to different subjects…