What is the ChatGPT bot and can it write your emails?
By now you probably have heard the hype about ChatGPT, a new AI tool that provides shockingly fluent and human responses to user questions and even writes passable poetry. What’s behind this new technology and could it write emails for you? Find out in our explainer!
You may have already interacted with a chatbot when you needed customer assistance on a website like an online retailer: You simply type in a question like “How do I get a return label” and these software applications give you a reply in natural-sounding language. So you can picture ChatGPT as a chatbot in “Ask me anything” mode. From “How do I change a car tire” to “What happens in the second act of Hamlet”, ChatGPT offers answers to users’ pressing questions in fluent phrases that sound like they were written by a human.
ChatGPT was developed and launched on November 30, 2022, by OpenAI, a San Francisco-based artificial intelligence (AI) company, and has already attracted more than a million users. The current release of ChatGPT is a beta version – improvements will be made to the software going forward based on public testing and feedback.
To use ChatGPT, you have to create a free account or sign in using your Google or Microsoft account. To use the chat box in your web browser, just go to the ChatGPT website and log in. Now all you have to do is type your question into the search field at the bottom of the page and press Enter – the chatbot will start answering you immediately. You can click “Regenerate response” to receive a differently worded answer. There is a navigation pane on the left side where you’ll see a list of your past questions, which you can return to by clicking on them. You’ll also find buttons for clearing the conversation and links to a FAQ.
The chat interface is extremely simple and intuitive to use, so the best way to get started is to simply log in and start asking questions!
Currently ChatGPT is completely free to use. As noted above, you will have to create an account and then sign in every time you start a new session. Your conversations with the chatbot are used to train the artificial intelligence, so in a way you are “paying” by agreeing to participate in the improvement of the product. You can give the response a thumbs-up or thumbs-down to show if the answer you receive is correct.
There seems to be no limit on the suggestions for how people are using ChatGPT, from writing jokes to answering homework questions. So why not try it for email writing? We tested the chatbot with the prompt “Write an email rejecting a job candidate before the interview stage“ and it generated an appropriate response with no spelling or grammar issues that could easily be customized using the candidate’s name and the position in question.
Then we tried a more detailed prompt: “Write a funny invitation to my family members to come for Thanksgiving dinner at 4 pm, I’ll provide turkey and the rest is pot luck.” This got us results we could simply copy-paste into the Compose E-mail window. So yes, you can use ChatGPT when you’re experiencing email writer’s block – just remember, the more information you provide, the better the results. And play it safe: Make sure you actually read the AI-generated message before you send it off to make sure there were no glitches.
But can ChatGPT get your relatives to RSVP?
ChatGPT is based on large language model (LLM) technology, which is used to accurately predict the next word or sentence in a text (autocomplete functions are one example of LLM technology). LLMs are made possible by machine learning, a kind of artificial intelligence in which algorithms are programmed to use training data to make predictions or decisions. In the case of ChatGPT, the machine learning technique is called Reinforcement Learning from Human Feedback (RLHF). It allows the software to provide responses that sound like they were written by a person because it was trained using massive amounts of data that really were written by people. OpenAI has said that when the product was being developed, humans “trained” the model by providing both sides of a dialog, the questions and the answers, to teach it what people expect a response to sound like. This was then expanded through huge amounts of data and information from the internet, e.g. sources like Reddit discussions and Twitter threads. This is what has given the tool its uncannily human style of writing.
OpenAI lists several things its chatbot can’t do. For example, it can’t answer questions based on information later than 2021. So it can’t answer a question like “Which NBA teams are expected to do well in 2023?” In addition, a question that is too general, like “How do I fix a leaky faucet?” will yield a reply that is also too general to be truly actionable – after all, there are many different types of taps and to fix yours, you would need how-to steps that apply to that specific model.
Many people have pointed out inaccuracies in the replies provided by the chatbot. For example, it was reported in early January that the tech site CNET experimented by publishing some articles generated using artificial intelligence and it was soon called out for factual errors in the posts. This should come as no surprise, as OpenAI specifically points out that users should always fact-check the answers. Indeed, it is important to realize that ChatGPT is not actually knowledgeable about the topics it writes about, but rather puts together an answer based on information that was first placed online by humans. So if the original data is incorrect, ChatGPT’s reply will also be incorrect.
Another significant limitation we noticed when trying out the chatbot is that it is often unavailable due to high demand. But this issue will presumably be solved once OpenAI have scaled up their systems.
One of the main concerns with ChatGPT seems to be that students will use it to cheat on their homework. Given the highly accurate answer the chatbot provided to our earlier question “What happens in the second act of Hamlet?”, we understand this concern! However, students are advised to exercise caution since, as noted above, answers are not always correct and can even be downright nonsensical. And ChatGPT users should beware – app developers are already working on tools that will identify whether a text was written using artificial intelligence.
Furthermore, while ChatGPT is programmed not to provide harmful or toxic responses, some fear that it may reflect and therefore spread human biases about things like race or gender. It could also be used to generate and spread misinformation, AKA fake news, at a speed much faster than humans could ever do working unassisted.
And many also worry that, like all new technologies, ChatGPT may displace human workers in the future – for example by generating advertising copy that is currently written by marketing professionals.
We hope you found this post on ChatGPT illuminating! Before you go to the GPT playground, why not leave us some feedback below!
