Reliance (ChatGPT is worse than smoking)

4/18/25

I’ve been thinking about reliance lately and have categorized my thoughts into 2 categories:

1. Human to tool

Day to day, I build web applications using Python, and recently the world has been taken by storm with tools like ChatGPT.

Reliance on these tools makes people dumber and is worse than smoking.

I wrote that pretty carefully, and didn’t write, “These tools make people dumber.” It’s the reliance that’s the problem.

While there are an infinite number of practical use cases for AI, these tools are conditioning the minds of well-intentioned people to think they’re getting value from the slop that comes out the other end.

I’m speaking from experience. I’ve seen simple features turned into completely unreadable spaghetti code that makes absolutely zero sense but “works”. Is that the point of software? Are we actually saving time by writing a feature in 15 minutes with AI that causes hours of refactoring later?

If it seems like I have a bad attitude about this it’s because I do. I’m not only frustrated because I have to deal with the consequences of poorly written AI code, but I’m also saddened that there are people falling in this trap and cooking their brains.

2. Human to human

Imagine building a web application that allows users to apply for a loan by submitting answers to a few questions.

You’re going to start building, but how do you handle the questions? You could:

• A. manually load an initial question set and then manage changes through a backend process requiring code changes
• B. Build a question administration page that allows business users to manage the questions themselves

A is bad and has high reliance. Any time the questions need to change, a developer needs to be involved. I’m allergic to this approach and have made a revelation about myself:

I want to be relied on as little as possible.

My objective function for building is:

min(reliance) s.t. the tool is useful

I’d proudly consider myself lazy in the sense that I want to write as little code as possible and strive to make others’ dependence on me as low as possible.

Obviously there’s a trade off between configuration and reliance (they’re negatively correlated). Building configuration takes time, but in my experience the world is a crazy place where everything changes all the time, so the investment in configuration is almost always worth it.

When I was a statistics tutor in college, I always told people that my goal is to teach general problem solving skills so they don’t need to come back. Not a great way to run a business but an excellent way to empower others!