The Dangers of AI

There are a lot of misconceptions about the dangers of AI right now. Most people think it’s dangerous in a way that it’s definitely not, and have no idea of the ways in which it actually is dangerous.

I’m in the physics program at NC State. I’m not an expert, but I’m experienced. I’ve trained a neural network.

First, let me assure you that you will not be left behind if you don’t learn how to use AI. That is marketing by tech companies, because the AI you have access to doesn’t solve any real problems. No one needed a website that was bad at math and also gave dangerous advice. Marketing has to convince you that there is a problem, and the problem they’ve chosen is that AI is going to take all the entry-level jobs. Is AI going to apprentice as a plumber? Is AI going to become a CNA and wipe people’s butts at the hospital? Which entry-level jobs are we talking about? It’s not even very good at coding, and that’s one of the things that it’s best at!

So that’s how it’s not dangerous. Here’s how it is dangerous.

When you hear about scientists solving problems with AI, they’re not using the paid version of ChatGPT. They have written the code, maybe even built the hardware, and trained the machine for years and even decades to solve a single problem. If you’re like most people, your entire experience with AI has been with large language models like ChatGPT. You’ve been intentionally misled to believe that they think. They don’t think. They just search the internet for the answer. If the answer isn’t out there, it will make up something out of words that has no connection to reality. You have to know enough about the problem to tell if the LLM is giving you inaccurate information.

How often do you think they do that? I can almost guarantee it’s worse than you think.

The best LLMs have a 30–40% failure rate, meaning that’s how often they hallucinate or provide inaccurate information.

I can’t stress this enough. When you use Google search, the first thing that comes up is an “AI overview.” One in three times you use Google Search, the first result you see is partially or entirely inaccurate. There’s no way to turn it off, so if you ever did care or notice that it wasn’t even a search result, you probably gave up after a while and started reading the top of the page because you just want an answer. Every third time you do this, it’s wrong. How many times have you used Google search today? If you’re not an expert in the field, and of course you’re not, that’s why you’re using Google, you are getting inaccurate information several times a day. Your brain is literally being programmed by the computer with erroneous information.

The good news is, you have power over the real danger of AI. Go to the library. Consult an expert. Look at a real website. Protect your brain.

Your fan,

Jonathan Byrd