My Attempt to Use ChatGPT to Produce an AWS SageMaker Image Classifier

My journey with ChatGPT through 15 prompts with 15 interesting answers

David Hundley
15 min readDec 14, 2022
Graphic created by author with AWS SageMaker icon in center

Needless to say, ChatGPT is the hottest topic in the AI/ML world right now. Many people have used it forU more benign cases to produce some very interesting results, so I thought I’d try to see if I could use ChatGPT in a “real world” business scenario. More specifically, I wanted to see if I could get ChatGPT to write me code that would produce an image classifier that I could use specifically with Amazon’s machine learning AWS service, AWS SageMaker. In a very short test I did via this tweet, I came into this experiment hoping it would go very well.

Spoiler alert: It didn’t.

To be fair, I didn’t walk into this experiment thinking it would go 100% perfectly. After all, I know ChatGPT is limited to only producing generic code, so I knew that I was going to have to make minor alterations to the code to fit a more specific, real world use case. The biggest problem I kept running into with ChatGPT is that it didn’t want to handle “larger” requests. For example, if I ask it to produce a full script for me all at once via a single prompt, it crashes like so:

--

--

David Hundley

Principal machine learning engineer at a Fortune 50 company, 5x AWS certified, 2x HashiCorp certified, 1x GCP certified, M.A. in Org Leadership, PMP, ChFC, CSM