Using GPT-4 to create "glass box" explainable models

Everyone is looking for ways to leverage the power of GPT-4 and other Large Language Models (LLMs) safely. But, many recognise that GPT-4 is really challenging to deploy in the enterprise. You cannot delegate responsibility to it to make critical decisions.

The model is not in your control, it can error and outcomes are not explainable (the “black box” problem).

Our focus is leveraging the power of LLMs (and GPT-4 specifically) to create models in Rainbird AI. Rainbird is a low-code graph-based knowledge representation and reasoning platform. We can now leverage GPT-4 to create powerful non-linear graph models that are “glass box”. The resulting ontology and probabilistic rules are subject to scrutiny, testing and amendment by experts with no coding experience. The resulting model can explain every judgement.

Rainbird is a very mature platform already used in enterprise for automating complex decision-making for organisations like Deloitte, EY etc. There are numerous use cases in financial services and healthcare.

Small video demo here: Rainbird Co-Author: Powered by GPT4 - YouTube

Comments welcome, and if anyone is interested in collaborating, please reach out.