ChatGPT, explained – The Verge

Some writers have declared that the debut of ChatGPT on November 30th, 2022, marked the beginning of a new chapter in history akin to the Enlightenment and the Industrial Revolution. Others have been more skeptical, wondering if this is just another overhyped tech, like blockchain or the metaverse.

What history will call ChatGPT remains to be seen, but here’s one thing I do know for sure: nobody has shut up about it since.

From injecting itself into presidential debates and Saturday Night Live sketches to creepily flirting with talking to you Her-style (well, briefly at least), ChatGPT has captured the public imagination in a way few technologies have. It’s not hard to see why. The bot can code, compose music, craft essays… you name it. And with the release of GPT-4o, it’s even better than ever.

Yet, as it gets smarter, the tech is also becoming less comprehensible. People are also getting more scared of what it can do, which is understandable given some are already losing their jobs to AI. It doesn’t help that a lot of sensationalism surrounds the subject, making it difficult to separate fact from fiction.

That’s why we decided to throw together this explainer so we can cut through all the BS together. You ready? Let’s begin.

Do you want the simplistic answer or the complex one?

The easy answer is that ChatGPT is a chatbot that can answer your questions by using data it’s gathered from the internet.

The complex answer is that ChatGPT is an AI chatbot powered by language models created by OpenAI that are known as generative pre-trained transformers (GPTs), a kind of AI that can actually generate new content altogether as opposed to just analyzing data. (If you’ve heard of large language models, or LLMs, a GPT is a type of LLM. Got it? Good.)

OpenAI is an AI company founded in December 2015. It created ChatGPT, but it’s also responsible for other products, like the AI image generator DALL-E.

No, but Microsoft is a major investor, pouring billions into the tech. Elon Musk co-founded OpenAI along with fired and rehired OpenAI CEO Sam Altman, Ilya Sutskever (who has since left), Greg Brockman, Wojciech Zaremba, and John Schulman. However, Musk eventually cut ties to create his own chatbot called Grok.

It will most definitely replace people with machines and — along with other AI bots like Amazon’s Alexa — basically take over the world. So you’d better start playing nice with them.

Nah, I’m messing with you. I mean, nobody knows for sure, but I highly doubt we’re going to see a job apocalypse and have to welcome in our new robot overlords anytime soon. I’ll explain more in a minute. 

Well, like I said, ChatGPT runs on GPTs, which OpenAI regularly updates with new versions, the most recent being GPT-4o. Trained by humans and a ton of internet data, each model can generate human-like conversations so you can complete all kinds of tasks.  

Where do I begin? The possibilities are practically endless, from composing essays and writing code to analyzing data, solving math problems, playing games, providing customer support, planning trips, helping you prepare for job interviews, and so much more.

Here’s just a short list of what it’s capable of: 

I mean, honestly, it could probably summarize this entire explainer. The AI world is your oyster.

Eh, not really. For all its hype, at its current level, ChatGPT — like other generative AI chatbots — is very much a dim-witted computer that sits on a throne of lies. For one thing, it hallucinates. 

Oh, sorry, not that kind of hallucination. Hallucination in the AI world refers to an AI-generated process in which the tool tries to extrapolate and create from collected data but gets it absurdly wrong, in turn creating a new reality. 

Honestly, I’m not a big fan of the word. It doesn’t really bear resemblance to actual human hallucinations, and I think it makes light of mental health issues — but that’s another subject. 

Incorrect information is a weak way of putting it. 

Sometimes ChatGPT actually fabricates facts altogether, which can lead to the spread of misinformation with serious consequences. It’s made up news stories, academic papers, and books. Lawyers using it for case research have gotten in trouble when it cited nonexistent laws.

And then, there are times when it gives the middle finger to both reality and human language and just spouts out pure gibberish. Earlier this year, for example, a malfunctioning ChatGPT that was asked for a Jackson family biography started saying stuff like, “Schwittendly, the sparkle of tourmar on the crest has as much to do with the golver of the ‘moon paths’ as it shifts from follow.” Which is probably the worst description of Michael Jackson’s family in the world.

Many AI researchers are trying to fix this issue. However, a lot of AI researchers think hallucinations are fundamentally unsolvable, as a study out of the National University of Singapore suggests. 

But hallucinations aren’t the only issue ChatGPT needs to iron out. Remember, ChatGPT essentially just regurgitates material it scrapes off the internet, whether it’s accurate or not. That means, sometimes, ChatGPT plagiarizes other people’s work without attributing it to them, even sparking copyright infringement lawsuits.

It can also pick up some really bad data. Likely drawing from the more unpleasant parts of the internet, it’s gone so far as to insult and manipulate users. Hell, sometimes it’s just downright racist and sexist.

Anything’s possible. But I would say that one thing is for sure: AI is here to stay, and so it wouldn’t hurt to learn how to leverage these tools. Plus, they really can make life easier in the here and now if you know how to use them. 

If you’re on a desktop, simply visit chat.openai.com and start chatting away. Alternatively, you can also access ChatGPT via an app on your iPhone or Android device.

Absolutely. The free version of ChatGPT runs on an older model in the GPT-3.5 series but does offer limited access to the newer and faster GPT-4o. That means free users, for example, will soon be able to access previously paywalled features, like custom GPTs, through the GPT Store.

ChatGPT also now freely supports the chatbot’s web browsing tool, meaning it can now search the internet in real time to deliver up-to-date, accurate results. The new model can also recall earlier conversations, allowing it to better understand the context of your request, while users can now upload photos and files for ChatGPT to analyze.

You do get more advanced capabilities through its paid tiers — ChatGPT Plus, ChatGPT Team, and ChatGPT Enterprise — which start at $20 a month. 

For starters, you have fewer usage restrictions, rendering them the better option if you plan on using ChatGPT often. Free users have usage limits OpenAI has yet to specify but has said that Plus subscribers are allowed to send five times as many messages as free users. The pricier Team and Enterprise subscription plans offer even fewer usage restrictions, though at this point, OpenAI has yet to divulge specifics.

Aside from being able to use ChatGPT longer, paid subscribers can do more. They can, for example, create their own custom GPTs and even monetize them via the GPT Store. Plus, only paid subscribers can access the DALL-E 3 model, which generates images from text prompts. 

Paid subscribers also get early access to the newest AI features. The voice capabilities OpenAI demonstrated onstage should arrive over the next couple of weeks for Plus subscribers, while ChatGPT’s desktop app for Mac computers is already rolling out for Plus users.

Custom GPTs are basically chatbots you can customize. There are millions of versions on the GPT Store that you can use to accomplish all kinds of tasks, from providing tech support to personalized hiking trail recommendations. Some customized GPTs currently trending include an image generating bot, a bot that makes logos, and a chatbot that helps people perform scientific research.

There have been some upheavals in the company — we’ll keep you in the loop.

Yes, there are quite a few, and each varies in terms of features, pricing, and specific use cases. One notable example is Google’s AI chat service Gemini. As a Google product, it offers deeper integration with Google services like Workspace, Calendar, Gmail, Search, YouTube, and Flights. The latest version, Gemini 1.5 Pro, also offers a longer 2 million token context window, which refers to the amount of information the AI model can understand.

Yeah! Did you know ChatGPT sounds like “chat j’ai pété” in French, which roughly translates to “cat, I farted.” Somebody even created a website with a cat who farts when you click on it, and I just can’t stop clicking.