What Is ChatGPT & How Does It Work?

If the news is to be believed, ChatGPT is set to take over the world, but what is it, and how does it work? Is it something that anybody can use, or is it reserved for only the brainiest among us? Let’s take a look at these questions and more.

What Is ChatGPT?

ChatGPT is an artificial intelligence, or AI. Before you start thinking about Data from Star Trek: The Next Generation or, worse yet, the Daleks from Dr. Who, the word “intelligence” in this particular case needs a bit of nuance.

ChatGPT can’t think for itself: if you present it with some information, it can’t form its own opinion about it like a human would. If you ask it something like “what’s the color blue?”, it blurts out some stuff from an encyclopedia, it won’t think of the first time it went to the beach or the color of the sky on a sunny day.

Instead, ChatGPT and programs like it scan the information they have and create an answer from there. For instance, in the above example it clearly used the dictionary definition about the color blue, which it likely got from the Britannica, Wikipedia or some other reference work.

This makes ChatGPT what’s known as a generative AI. It can generate answers from existing material. It can’t, however, do its own thinking. Ask it an opinion, and it’ll just regurgitate one it was fed earlier, it can’t create connections between two or more sets of facts and draw its own conclusions. Usually, it’s pretty honest about it, too.

As a result, you could argue, and plenty of academics have, that calling ChatGPT an artificial intelligence is misleading. OpenAI, the company that created it, calls it a large language model, or LLM, while others have suggested calling it a probability calculator or a statistical model of some kind; arguments for a name-change abound.

For what it’s worth, GPT stands for generative pre-trained transformer, which should give you an idea of what the people who designed thought it could do: generate content from a set of data it had been fed, transforming it somewhat.

What Can ChatGPT Do?

Downplay it as much as you want, it’s still impressive what ChatGPT can come up with. For example, in our article on how to use AI for school, we go over how it can be used to write an outline for pretty much any kind of academic paper, and you can even get it to spit out a poem in the style of your favorite poet.

You can get it to write entire blog articles (which is causing some issues), write novels, put together marketing campaigns; you name it, ChatGPT can do it. As a result, articles and opinion pieces about how writers, journalists, lawyers and plenty of related professions abound — it seems the only job that’s safe right now is writing articles about how ChatGPT spells doom for this group or another.

Admittedly, seeing ChatGPT work is pretty impressive, at least at first: you prompt the bot — or one of its many alternatives — just about anything and within seconds, or minutes if you had a particularly tall order, it spits out more or less what you asked it to.

However, once the novelty wears off, holes start appearing. For one, ChatGPT’s output always feels a little stale: it’s not the fresh kind of writing you’d get from a real talent, rather the forced scribbling from a second-rate imitator. It also repeats certain stylistic trappings and overall feels very stiff and unnatural — even if you use the newer GPT-4 model instead of its older sibling, GPT-3.5.

What also doesn’t help is that much of GPT’s replies won’t be up to date: it was trained on facts up to September 2021, so if you ask it anything more recent, it can’t help you. As a result, much of its information on politics and cutting-edge fields is woefully inadequate.

Speaking of which, we’ve mentioned training a lot so far, without really going into it. In this case, training consists of taking the idea and “feeding” it lots of information. There are several ways to do this, but in the case of ChatGPT the method used is called unsupervised learning. Basically, the AI was supplied with a whole bunch of information from the web and told to memorize it.

Note that at no point has GPT been connected to the internet: the information fed to it was static, so to speak. Since we have no idea what may happen if we let a generative AI loose on the internet, it seemed the developers prudent to keep it quarantined.

Of course, ChatGPT wasn’t just fed the web and then declared finished; training also involved having it communicate with researchers who curated its answers. There was no way OpenAI would let another Tay happen, a chatbot by Microsoft that degenerated into a racist maniac in just a few hours of interaction with social media.

As a result, ChatGPT is often described as being “safe,” meaning that if you ask it a laden question it will just give a neutral response. That is, if it responds at all; the further you go into dodgy territory, the bigger the chances that the bot won’t answer, or remind you that being a racist is not cool.

ChatGPT Accuracy

However, despite its training, there are still some issues with ChatGPT, the biggest being accuracy. Not only is its information a little out of date, if it doesn’t know an answer it will make one up.

This bears repeating: AI will straight up lie to you if it’s not sure about an answer. These hallucinations, as they’re known, are fairly common, too, so you need to be careful when using ChatGPT.  We wouldn’t trust any of its output, you need to verify facts for yourself and make sure what you were given is correct, the AI can’t do it for you.

Using ChatGPT

With all the development that went into it and all it can do, you may be surprised to find out that using ChatGPT is very easy. All you need to do is visit the OpenAI website, and go from there. Before you know it, you’ll be viewing the chatbot itself, wondering what to type.

Since you can ask it anything, pretty much, we won’t tell you what to do. However, there are a few things to keep in mind. For one, the broader the question, the more nonsense ChatGPT will spew at you. Kinda like a drunk dude at a bar, ChatGPT will keep on rambling if you give it the slightest sliver of a chance, so make sure your queries are to the point.

If you want it to write a blog post, define exactly what you need it to do. If you need an outline for an academic paper, be specific about the subject. In our experience, it’s best to have AI augment your natural skills and abilities rather than have it actually do the work for you. Between hallucinations and derivative output, it’s in no way a labor-replacing device.

It is, however, a very good labor-saving device: we go into details on all the ways you can use AI to improve your writing, for example. It’s also great for students, lawyers, and anybody else that could use a hand when working.

Also, here’s a final tip: though ChatGPT will tell you you don’t need to be polite when asking it things, we feel it’s a good idea to be nice to it, just in case that robot uprising does come about.