My Inner Julia Child

I have a confession to make: I’m a huge Julia Child fan. I grew up watching (and attempting to follow along with wildly varying degrees of success) her French cooking show on PBS. This spawned a lifelong love of cooking and of experimentation when cooking. I love making recipes my own with a bit of a tweak here or addition there.  

But I can’t improve a recipe if I can’t make it the way it was originally intended and if I don’t understand why it works that way.  Everything starts with the original recipe.  

Foundation models like OpenAI’s GPT LLMs are the original recipes of AI.  

I asked ChatGPT to explain Foundation Models to an 8 year old using recipes as an analogy.  This is what it generated:

“An AI model is like a recipe for making something, but instead of making a cake or cookies, it’s a recipe for making a computer program that can do things like recognize pictures or understand language.

A foundation AI model is like a special recipe that someone else made and shared with us. It’s like using a recipe for making chocolate chip cookies that someone else already tested and perfected, instead of making up our own recipe from scratch.

Just like how we can add our own special ingredients to a recipe to make it our own, we can add more information to the foundation AI model to make it even better at understanding things or recognizing things. It’s like adding sprinkles or nuts to our cookies to make them even more delicious!”

(I shared this with my 8 year old daughter and she both understood the analogy and wanted me to make cookies with her.)

While OpenAI’s GPT models are probably the most well known Foundation Models that are Large Language Models, there are hundreds of others. 

What Foundation Model boodleGPT Uses and Why

We are building boodleGPT using OpenAI’s GPT-3.5 as its Foundation Model (we’ll eventually move to GPT-4).

We selected OpenAI’s LLMs because of:

1. Ease of development (which leads to speed to market)
2. Size of community. 

However, as we’ll explain in a future post, we’re doing more than just taking GPT-3.5 and adding our proprietary insights to it (That would just be TOO easy!).

.

_{References:
Reflections on Foundation Models (stanford.edu)
What Are Foundation Models? | NVIDIA Blogs
What are foundation models? | IBM Research Blog
What Are Foundation AI Models Exactly? — ITRex (itrexgroup.com)
What are Foundations Models in AI? – Definition from Techopedia
AI Explainer: Foundation models ​and the next era of AI – Microsoft Research}