Skip to content

An Unexpected Friend in a Grueling Journey–ChatGPT

By: Camey VanSant

By Sharad

August 2023 – Nov 2023

Hey Everyone,

In a happy conclusion of my months-long research internship, our lab’s paper has finally been accepted to be published in a peer-reviewed medical journal and will be presented at the COSM (Combined Otolaryngology Spring Meetings) 2024 annual conference. While the project has been incredibly interesting and stimulating, it wasn’t without its fair share of challenges. As I reflect on the past months, I feel it’s the perfect time to talk about the seemingly insurmountable challenges I faced and how I overcame them.

For roughly the first 4 months of my gap year, I had been treading through a forest of Stack Overflow (an online forum for coders) and random documentation manuals but ended up more lost than when I started. I had been looking for something that could help me upgrade my surgical tool tracking system from an older, deprecated version of tracking to a newer one. (To catch you up to speed, I’ve been working in a research lab to develop an intra-surgery feedback system to help surgeons conduct safer surgery.) However, everything I could find was either too outdated or way too complicated to implement. So, after hours spent trying to learn from others online, I turned to someone (or rather something) that had consistently been making headlines – ChatGPT.

My very first interaction with ChatGPT 🙂

ChatGPT provided not just answers but a collaborative brainstorming experience. It patiently guided me through the maze of upgrading my code from TensorFlow 1 to TensorFlow 2. (TensorFlow is an AI tool that you can use to track objects in an image or video.) Unlike other online forums, I could feed ChatGPT my entire project so that it could tailor-make recommendations for my specific use-case. The project proceeded smoothly for a while after that. That is, until I got another message from the grad student supervising my project – scrap the entire TensorFlow software and transfer it to YOLOv8 software, a better overall object-tracker. This time I went straight to ChatGPT but was shocked to find that alas it could not find a fix for everything I wanted. So, I went to Bard – Google’s chatbot – and by comparing the results I got from both places, I was able to stitch together a working system.

All in all, I’m glad I had this experience as it changed my long-standing views regarding generative AI and chatbots in general. Previously, I used to think chatbots’ capabilities were mainly for show and didn’t have a lot of real-world applications. But now that I’ve seen them actually generate code on their own, I’m way more attuned to all they can do and will be more than willing to call upon them if the need arises in the future 🙂

A snapshot of my tool-tracking model in action along with some of the code used to build it
Categories: Sharad