AI, LLMs and beginners

This AI thing has been going on for a while, specially the LLM part of it. I understand why there is hype for it, specially from VCs, and mostly from people who *checks notes* are not in the high-techs.

My students are using a lot of ChatGPT (and the others too) and I keep telling them to not use it, not because I don’t want them to use LLMs at all, but because LLMs suck. They are just an interface to a computer, and if you’ve ever done computer programming, you know that a computer does what you tell, not what you mean.

As a beginner (in Software Engineering, System Administration, etc) you still don’t know what you want a computer to do, that’s why you tell a program what you mean, instead of what it should do. We can see this problem everywhere. Here’s a real-life example from today.

I’m using the nginx web server, I’d like to allow only the domain example.com, reject everything else

What my student meant, is that, if you access the nginx web server via an IP address, then it should show nothing, if it’s a specific domain, such as example.com, then it should show the web page.

What ChatGPT understood is about access control and suggested the following

location / {
            root   /usr/local/www/nginx/;
            index  index.html index.htm;
            allow example.com;
            deny all;
        }

As a beginner, my student thought “well, that was easy!”, and then he kept wondering why he can’t access his web server, for 2 days.

And that is why you should not use ChatGPT (or any kind of an LLM) as a beginner.

As soon as you understand how a computer works, then go on, use whatever you want. Hell, even use JavaScript. But before using ChatGPT or JavaScript, please learn how a computer works first.

Reply via email.