Return to site

What is the GPT in chatGPT?

It's as simple and complex as cooking...

January 30, 2023

Ask someone if they've heard of chatGPT and you'll probably get an enthusiastic nod and an amazing example of a random paragraph it generated for them. Then ask them if they know what GPT stands for and you'll most definitely get a blank look. So they asked me if I knew and I told them the GPT stands for Generative Pre-Trained Transformer. The inevitable follow up question is how does it work? Cut to my blank look.

Of course I'm no expert in this area so I started researching. After reading a few choice picks from a Google search ranging from the Wiki page on GPT to a Cornell University paper on Attention Mechanisms, I thought wait wouldn't it be easier if chatGPT to explained itself. So I asked if it could explain GPT simply using a cooking analogy hehe. Here is the response.

A good cooking analogy to use when explaining GPT is to think of it as a chef that has learned how to make a variety of dishes by observing and experimenting.

 

Just like a chef learns the ingredients, flavors, and techniques used in different dishes, GPT learns the patterns, relationships, and styles in text through its training. The more a chef is exposed to different dishes, the more they learn and can make a wide variety of dishes on their own. Similarly, the more text data GPT is trained on, the better it becomes at generating coherent and fluent text.

 

Now, when you give the chef a recipe to make, they use their understanding of ingredients and techniques to make the dish. Similarly, when you give GPT a prompt, it uses its understanding of patterns and relationships in text to generate a response that is similar in style and content to the text it has seen during training.

 

In this analogy, the training data for GPT is like the ingredients and recipes that the chef has seen, and the prompt is like the recipe that you give the chef to make. The output of GPT is like the dish that the chef has cooked, it uses its understanding of text to generate a response that is similar in style and content to the text it has seen.

So with this analogy, "restaurants" will be the different applications that the market will create and by the looks of things there will be some amazing, three Michelin starred ones. Each with their own unique flavour so to speak.

Here at CloudE8 our restaurant is called O.LLA. Come by and try some of the dishes we've trained our chef to create. With many more to come. Yum.