My First Encounter with TDD

Reflecting on my early days with ChatGPT brings back memories of my inaugural programming course in Fortran. Although I found programming manageable, debugging was a chore I loathed. It was disheartening to resolve an issue only to be met with compilation errors stemming from minor syntax mistakes. Cleaning up code post-problem-solving was never enticing, and I regarded programming as a necessary hurdle to complete my degree.

Despite the increasing coding demands over the years, my enthusiasm for programming did not grow. It wasn't until my first job that I discovered my talent for algorithm design and selection. I relished the opportunity to experiment with algorithms, dedicating much of my time to this pursuit. Yet, my feelings toward coding remained unchanged until my workplace adopted TDD.

Initially, I felt compelled to adapt to a new workflow that differed from my established practices. Traditionally, I would write tests after ensuring my code functioned correctly through manual checks. However, after attending a TDD lecture and embracing a new mindset, everything transformed.

The lecturer emphasized that TDD is fundamentally a dialogue with your computer. Writing a test is akin to articulating your goals and seeking assistance from your machine to achieve them. In essence, creating a test is a declaration of intent.

This shift in perspective eradicated my aversion to debugging. I began to enjoy my interactions with the computer, viewing it as a collaborator in my endeavors. This experience ignited my fascination with enhancing communication with machines.

To me, ChatGPT represents an evolution of TDD. At its essence, TDD serves as a straightforward protocol for two participants—human and machine—to engage until a target is met. ChatGPT extends this dialogue, enabling the pursuit of diverse objectives.

Section 1.1: Leveraging ChatGPT for Work

My initial exploration involved assessing how ChatGPT could streamline my work processes. My first endeavor was to see if it could help with mundane writing tasks, starting with the creation of a job posting.

ChatGPT's assistance in drafting a job description saved me significant time that I would have otherwise spent scouring the internet for examples. It provided a well-structured format, which I then tailored to include specific details about my organization and team.

Section 1.2: Crafting Interview Questions

I then tasked ChatGPT with generating interview questions. While it provided a solid foundation, I sought to incorporate more context, including technology concepts related to my organization and soft skills such as communication and leadership.

Before long, I was engaging with ChatGPT about optimizing our processes to minimize bias and enhance fairness in hiring practices. While I could have pursued this information through research or HR, the insights I received from ChatGPT were impressive and informative.

Getting Personal: Exploring Beyond Work

While my initial focus was professional tasks, I eventually ventured into personal projects. As an enthusiastic writer, I sought ChatGPT's assistance in overcoming writer's block by generating blog topics. I even requested a 5000-word blog based on one of the ideas, along with ten engaging titles.

One restless night, I turned to ChatGPT for home entertainment suggestions. Following its advice, I enjoyed a bubble bath while engrossed in a book, providing a refreshing escape.

On Valentine's Day, my wife requested surf and turf for dinner, presenting a challenge since I had never cooked lobster. Relying on ChatGPT for guidance, I received comprehensive instructions for preparing the lobster, which turned out to be a delightful success!

Testing ChatGPT's Technical Abilities

After numerous successful non-technical prompts, I decided to challenge ChatGPT with technical coding tasks. My first challenge was to scrape data from SharePoint Online, an area where I had no prior experience.

ChatGPT not only generated the code but also provided detailed instructions for connection. I was impressed with the code it produced!

import requests

from requests_ntlm import HttpNtlmAuth

# SharePoint site URL and API endpoint for the list you want to scrape

api_endpoint = "/_api/web/lists/getbytitle('your_list_name')/items"

# Authentication credentials for accessing SharePoint

username = "your_username"

password = "your_password"

domain = "your_domain"

# Set up the API request headers and authentication

headers = {

"Accept": "application/json;odata=verbose",

"Content-Type": "application/json;odata=verbose"

}

auth = HttpNtlmAuth(f"{domain}\{username}", password)

# Send the API request to SharePoint and get the response

response = requests.get(site_url + api_endpoint, headers=headers, auth=auth)

# Check if the API request was successful

if response.status_code != 200:

print("Error: Failed to retrieve data from SharePoint.")

exit()

# Parse the response JSON and extract the relevant data

data = response.json()["d"]["results"]

for item in data:

print(item["Title"]) # Replace "Title" with the name of the column you want to extract data from

I later asked ChatGPT to create a weighted Jaccard similarity for string matching. However, it failed to meet my expectations, applying weights incorrectly. I returned to basics and requested the standard Jaccard similarity code, which was mostly correct, with only minor errors.

As I provided more context, ChatGPT began to yield more relevant suggestions for calculating a confidence score based on a nearest-neighbor methodology.

Initial Insights: Learning from Experience

This experience taught me valuable lessons. Firstly, utilizing ChatGPT improved my communication skills, requiring me to articulate my needs more clearly. This exercise can enhance my interactions with colleagues, as I often assume they share my level of understanding.

Secondly, I recognized that my vague initial requests stemmed from my own uncertainty regarding what I wanted. When ChatGPT's responses missed the mark, it prompted me to refine my queries.

This revelation was transformative, reminding me of my early experiences with TDD. Previously, I viewed TDD as a means for computers to assist in iteratively achieving predetermined outcomes. However, engaging with ChatGPT highlighted that the computer can also help refine and clarify my ultimate goals.

Addressing Potential Pitfalls

Despite my numerous positive experiences with ChatGPT, my exploration of string matching raised concerns about Gell-Mann Amnesia—a term coined by Michael Crichton that describes a tendency to overlook inaccuracies in unrelated fields after recognizing them in one's area of expertise.

Initially, I trusted ChatGPT's outputs in unfamiliar territories, but I grew more critical when seeking assistance within my domain. This realization prompted me to approach information cautiously across various fields, understanding that if the responses were unreliable in my area, I should be skeptical elsewhere.

Establishing Safeguards in the Workplace

After just a few days of using ChatGPT, I became convinced of its potential benefits for my team. However, I recognized the need for safeguards to mitigate potential risks.

To protect sensitive company data, I advised my team to refrain from using ChatGPT on our corporate network until it had undergone an information security review. While individuals could use it at their discretion, I stressed the importance of ensuring its safe implementation within the company.

Additionally, I prohibited the use of company information in ChatGPT. The risk of sensitive data exposure is significant, especially since ChatGPT is continuously trained on user inputs.

Lastly, I imposed a ban on using ChatGPT to evaluate any human-produced material. Given the biases inherent in AI models, it is crucial to avoid relying on technology for evaluating candidates until we can ensure fairness and equity.