Learning Faster, Smarter

Image Credit: Pexels

What is GenAI for me?

As a software engineer and machine learning enthusiast, I frequently work with Python, Java, React, and even CUDA code. Over the past few years, Generative AI (GenAI) has become an integral part of my everyday routine. As a computer engineering graduate student at NYU, my curiosity about how technology works, drives me to explore the power of GenAI. I first came across tools such as ChatGPT during my undergraduate studies. I have personally experienced how it has grown in its capabilities in the past couple years. GenAI for me is not just a technology but a tool that helps me for my personal productivity by assisting me in writing reports and working on complex technical implementations to support my academic and professional journey.

How do I use GenAI for my studies?

I frequently use ChatGPT to learn new topics, summarize lengthy research papers, break down complex concepts, write boilerplate codes, and even refactor code to improve performance.

During my second semester at NYU, I had taken an advanced course on Deep Learning by Prof. Chinmay Hegde. This course was slightly difficult for me, as I didn't have much experience in the subject. One of our projects involved generating adversarial examples to attack a pre-trained ResNet-34 image classifier. We had to test out different techniques to reduce the accuracy of the model, yes, you read it right, reduce the accuracy and not improve it. Turns out, reducing a model’s accuracy is more difficult than improving the accuracy. The use of GenAI was permitted by the professor with proper acknowledgements. For these projects, I had to do deep research and understand the core concepts before starting the project. I used GenAI to learn about them and help me write the basic structure for these attacks. I would prompt GenAI with a very specific question such as, "Can you explain how Momentum PGD works to attack a ResNet-34 model and provide a well-commented PyTorch implementation that generates adversarial examples using it?" and it would give me a detailed explanation on the topic along with a well written code. I used the given code and played with the different parameters by understanding the math behind them using GenAI.

Due to GenAI, developing a basic understanding of topics I had never worked on became quite easy and quick. Specifically, I found it helpful that ChatGPT excluded information that was less relevant. For example, earlier, when I had to study a new topic, I would read the official documentation, watch YouTube videos, and read research papers, blogs, Stack Overflow threads, etc., just to understand what the concept was. This used to take me hours just to filter out content which was relevant to me for the use cases I was actually learning it for. Finding answers to a specific question sometimes took even more than an hour. This part of learning has become quicker because of ChatGPT, as it only highlights the specific answer I am looking for from all these resources. ChatGPT is continuously improving everyday with newer and finer models and I do agree, it still does hallucinate sometimes and give out incorrect answers, but it’s still faster as compared to the traditional method of learning.