ChatGPT (Chat Generative Pre-trained Transformer) is a natural language processing chatbot powered by the GPT family of large language models. It was launched on November 30, 2022, quickly gaining millions of users within days.
ChatGPT is now one of the most widely used online tools in the world, with businesses finding new ways to put it to work on an almost daily basis.
But what is ChatGPT, what’s it primarily used for, and how does the chatbot actually work? We’ll answer all of these questions and more in this guide.
What Is ChatGPT?
ChatGPT is a generally available natural language processing tool that has been powered by the GPT family of large language models since its release. Language models are probabilistic models that predict word sequences based on training data. Large language models (LLMs) use deep learning techniques to reach higher levels of linguistic understanding and capability.
GPT stands for Generative Pre-trained Transformer. GPT models are pre-trained on huge corpora of unlabeled text data. What is learned from these datasets is then used by ChatGPT to generate responses to queries from users in real time.
ChatGPT is owned and maintained by AI startup OpenAI. Microsoft has invested around $13 billion in the company since 2019. Since its release, a number of ChatGPT alternatives have been launched, including Google’s Bard and Anthropic’s Claude.
Now you’re clued in on what ChatGPT is, here are a few facts about the chatbot that illustrate just how popular it has become.
- More than 180 million people used ChatGPT in August 2023
- ChatGPT receives around 10 million daily queries
- OpenAI spends around $700,000 a day to keep ChatGPT going
- 2 million people in the US pay for ChatGPT Plus
- ChatGPT became the fastest-growing platform in history after it launched*
*This record was later broken by Meta’s Threads app.
ChatGPT is powered by the GPT family of language models. When it was initially released, it used the GPT-3 large language model to generate responses. Now, the free version of ChatGPT is powered using GPT-3.5, while the paid version is powered using GPT-4.
As we’ve covered, GPT stands for Generative Pre-trained Transformer, which is a type of Large Language Model. GPT models are trained on a corpus of text data that includes internet resources like webpages, documents, files, books, and stories.
GPT-4 has more “parameters” than GPT-3.5. Parameters are adjustable components of large language models that play a central role in transforming inputs into outputs. The more parameters, the larger and more powerful the model is.
The GPT models use finely tuned, specialized algorithms to look for patterns and sequences in the training data – the underlying structures that exist in all written text.
This is an example of “non-supervised learning” – which is the kind of learning that ChatGPT does. When it is trained, inputs (i.e. text to respond to) are not associated with any specific outputs. This is why ChatGPT is better at understanding the intricacies of human thought, speech, and written messages unlike, say, a “help” bot you might find on a website.
For comparison, these customer support chatbots – which are simpler and more limited – use a method called “supervised learning” – where inputs and outputs are tied together. This is why they appear more mechanical and don’t answer questions cogently anywhere near as often as ChatGPT does.
Although ChatGPT is extremely capable and useful thanks to its complex training processes, they’re not perfect, nor is ChatGPT powered by a human mind.
Logical and factual inconsistencies are common, as is the generation of false information. Instances of this are commonly referred to as hallucinations. Some requests that pit different parts of ChatGPT’s logical infrastructure against one another can also lead to strange outputs being generated.
This is a simple explanation of an incredibly complex process, but at its core, that’s how ChatGPT works. Due to the fact that the human experience is full of biases, ChatGPT will and does exhibit some biases as it picks apart the underlying structures that written text is based on.
This is one of several ethical issues surrounding the continued use of artificial intelligence.
Posing an ethical question to ChatGPT.
ChatGPT’s use cases are almost endless – people have been using it to complete all sorts of work. Some of its most popular use cases include:
- Summarizing documents and files
- Paraphrasing complex text
- Generating teaching materials (e.g. questions)
- Creating formulas for spreadsheets
- Coding duties in various programming languages
- Analyzing and interpreting data
- Brainstorming and generating ideas
- Proofreading (but not fact-checking*)
- Generating content (articles, product descriptions)
- Reviewing legal and financial documents
- Customer support/service (API)
ChatGPT is much easier to use if you use specific ChatGPT prompts, which elicit specific, useful responses from ChatGPT because they’re designed to minimize confusion.
*ChatGPT is trained on data in a set that was compiled before 2021. This means that it struggles to give accurate answers about events that took place after this time.
Generating formulas with ChatGPT.
If a website is claiming to give you access to ChatGPT outside of the OpenAI domain, treat it with extreme caution. As with all popular websites and tools, it didn’t take long after ChatGPT’s launch for scammers to start to impersonate it.
ChatGPT and AI scams are now extremely common, so it’s crucial you keep your wits about you when using these sorts of tools online.
If you want to sign up for ChatGPT, you’ll need an email address and your phone number to log in. You’ll also need to create a strong, unique password, which is particularly important if you’re going to be using ChatGPT to review things like confidential documents.
OpenAI requires you to hand over your mobile phone number because it stops people from just constantly making accounts with new email addresses. However, once more, it’s also worth bearing in mind that you can’t really use the service anonymously – another factor worth considering when you use it.
Yes – ChatGPT now has an official app for Android and iOS, so you can use the chatbot on the go. The app is rated 4.7/5 on the Google Play store and 4.9/5 on the Apple Store.
There are lots of fake ChatGPT apps floating around these marketplaces, so be careful that you’ve definitely got the official app in front of you before you download anything.
Yes – there’s a free version of ChatGPT that’s been available since the November 2022 launch. You can get chatting by signing up in the way described above. The free version of ChatGPT is powered by GPT-3.5.
There is also a premium version of ChatGPT called ChatGPT Plus. It costs $20 per month to use, but it utilizes GPT-4 rather than GPT-3.5, which is a lot more powerful. GPT-3.5 has over 175 billion parameters, but GPT-4 has 1.7 trillion.
OpenAI will now also provide developers with the ChatGPT API, which means they can incorporate ChatGPT into their products, services, and business infrastructure.
Planning a trip to Athens, Greece, with ChatGPT's free version.
Throughout this guide, we’ve made reference to a few of the “problems” that are baked into ChatGPT’s mechanisms, as well as the fact many scammers are hopping on the AI bandwagon and using it to con unsuspecting targets through various means.
There are other things worth considering firstly when using ChatGPT, and secondly when looking for AI tools for other uses online.
When using ChatGPT
If you’re using ChatGPT at work, always be transparent about your usage with your manager and follow your business’s AI guidelines if they exist. Always proofread work created by ChatGPT, especially if it’s for public consumption or being sent to clients and customers.
Ensure that you’re given the all-clear to input certain types of information for ChatGPT to review, and we’d recommend deleting conversations that include personal or sensitive information once you feel the chatbot has served its purpose.
When looking for other AI tools online
If you want to use another AI tool, conduct extensive background research. As we’ve said previously, there are quite a few apps on the app store that claim to be useful AI tools, but are severely limited and will coax you into paying hefty subscription fees after only a short period of use.
Google’s Bard, Anthropic’s Claude, and Perplexity.ai are three alternative chatbots that you can use to get started and are perfectly safe. You might also find Grok vs ChatGPT an interesting new chatbot clash to consider. But if you’re hooked on ChatGPT like the rest of the world, then you may not have to.