A Dip Into Generative A.I.

Taking in the sunset at UC Davis

"A journey of a thousand miles begins with a single step"
- Lao Tzu

I know I’m late with this post—finals week was a blur that left me with little to no time for personal endeavors. But now I’m back! It’s been a week since my last final, and I’ve made relatively good progress in exploring Generative AI. I finished up the introductory course, albeit it was only 30 minutes, but it was a lot to digest. Not only that, I also completed the next course in the pathway, Introduction to LLMs.

Unpacking it all, Generative AI is as it sounds, an AI that generates things. It is a subset of Deep Learning, meaning it's general architecture mimics that of a human's neural network. These Generative AI models are trained in one of 3-ways:

1) Unsupervised - where the data input is not labeled.

2) Supervised - where the data input is labeled and the model optimizes based off of error each use.

3) Semi-Supervised - a good mix of both techniques above.

Once trained, these models take input in the form of a prompt. These prompts are processed through a transformer, where they pass through an Encoder and Decoder.

Simply put, the Encoder takes the prompt, filters out noise and emphasizes key points before passing it to the Decoder. The Decoder then uses this refined input to create a target output sequence.

As mentioned earlier, Generative AI is used to create new content. This means it has applications relevant to any industry. The potential use cases are endless, hence the recent surge of interest in the field.

Many companies have developed Foundation Models, which are promoted as a pre-trained model with a vast amount of fundamental knowledge. The goal is to provide an easily accessible pathway for any company to leverage generative AI.

I find these Foundation Models to be particularly exciting as it brings what is considered a very complicated, technical field to the doorstep of any company. With proper product support, they have the potential to streamline work processes and serve as a source of inspiration for product development.

As I move into the upcoming week, I aim to progress further down the Beginner: Introduction to Generative AI Learning Path. Having completed the introductory course as well as Introduction to LLMs, the next course on the chopping block is Introduction to Responsible AI followed by Prompt Design in Vertex AI. I am greatly looking forward to this next week as I will be delving into the intricacies of model design and utilization.

Until the next step...

~ Cyrus Foroudian