I would like a ‘Seinfeld’ to go please…

In the past I had written about using Generative Adversarial Networks to create novel content in the form of voice, prose, poetry, art etc. With the progress made in the field of Generative AI, the process to make this happen has been even further simplified.

When I read the book “Life 3.0” by Max Tegmark, I was a partly excited and partly nervous by the possibilities of what artificial general intelligence (AGI) can do in our lives in the near future.

There has been a flurry of activity with interesting products built with Gen AI that can aid with creation of content. One of the use cases I read in Max’s book is possible with the tools we have today.

Picture this hypothetical scenario:

1. You create a storyline for movie or a TV series or your own Youtube channel that appeals to a pretty broad audience. You will of course know this ‘appeal’ by leveraging the viewership analytics of ‘what’ gets the most eye traffic across all the media outlets.
2. With the analytics you have available, you create and fine tune every character in the story — their looks, clothing, voice, accents, mannerisms to name a few
3. As this gets shared on a public media outlet, you get to use the audience feedback to fine tune the course of the story

You just need the framework to leverage the different options for Large Language Models (LLMs) to build/fine tune/and optimize this content.

Luckily for us, there are few early options of creating interesting pipelines and agents to do such work that wrote about earlier here

The JARVIS framework from Microsoft makes this possible by creating a robust pipeline with feedback control built in for such efforts.

It takes in a user input to breakdown the natural language query into different tasks, pick the right LLM model for creating text, audio, video content, execute the models, and stitch the response together.

When we have such a model working to churn out novel content, the possibility depicted in Max’s book is not too far in the future.

There is a surge in the number of products (MidJourney, Stability Diffusion, Runway to name a few) that create novel content in one or more mediums. It is exciting to figure out ways to stitch the ones that are open for API integration and create your own custom pipelines.

For fun, I wanted to see if I can create a simple comedy teleplay script. This could be a first step before I create the characters, define their looks/clothing/personas, and stitch them into a video production. I have to say, there are parts of it which are funny. I am a fan of the creative mind of Larry David, so I stayed with a classic. Here is an output I got from ChatGPT. It has the quirkiness and the dry humor from the original characters (which of course the model has learnt with the corpus) — I thought it was funny!

Traditionally, there has been a fine line between left brained and right brained individuals who bring their expertise in analytical and creative domains respectively.

Thanks to the available options with the GenAI products, tools, and frameworks, this fine line becomes less apparent. As someone with a strong left brain, I am excited about these possibilities.

Stay creative (all my fellow quants and engineers)!