At the end of November 2022, the world was stunned when the conversational AI chatbot, ChatGPT, was launched to the public. Here was a bot that could write intelligently on demand, answer all sorts of questions and even tackle the deeper matters of meaning and ethics. In a mere month, ChatGPT had already garnered over 1 million users.
What is ChatGPT?
In essence, ChatGPT is an AI chatbot that can conduct meaningful two-way conversations with humans via a conversational text interface that are remarkably natural-sounding as if you’re talking with a human.
It is a Generative Pre-Trained Transformer (GPT) that uses the GPT-3.5 Large Language Model (LLM) that was developed by OpenAI, which was originally sponsored by Elon Musk and Sam Altman (among others) as an open-source, non-profit research foundation in 2015.
ChatGPT basically learned from what was on the Internet pre-2021: websites, Wikipedia, books, etc. and continues to learn based on Machine Learning with human feedback (Reinforcement Learning from Human Feedback (RLHF))
Generative AI uses algorithms comes up with new outputs, like content, code and images, that are derived from the huge set of data that the AI learns from.
What can ChatGPT do?
Drawing from the GPT-3.5 LLM, ChatGPT can produce answers to a myriad of questions and present them in a succinct piece of text. It can write articles, theses, summaries...the list is endless... in any tone or ‘voice’ or level that you specify. For example, when asked to explain 'how does an aeroplane fly’, it replied:
When asked to do the same for a 10 year-old, it came up with:
Coders are finding that ChatGPT can solve complex coding problems with great accuracy, and other disciplines are also using ChatGPT to come up with solutions as they play with questions and pose problems to push the bot’s limits. All this ‘play’ is part of Machine Learning and continuous feedback.
What can’t ChatGPT do?
Firstly, it was trained on data from 2021 and before so anything after that is not accessible.
Nor is any live data available, such as breaking news or theatre times.
And what about inspiration, creativity, and innovation? Yes, to a degree ChatGPT can ‘create’ new things based on learnings from what already exists, but can it come up with something entirely new? Also, can it really understand and generate complex language such as the nuances of humour?
Where can ChatGPT be used?
Some people see ChatGPT as a new kind of search technology that answers your search in one piece rather than giving you a choice of links as in the case of Google. However, this is limiting as it cannot answer things like, “What is the weather today?” or give real-time sports results and news.
Content generation is the stand-out area where this technology can be used since it is so effective at generating well-written copy, especially if it’s used to create a basis for copy that can then be edited.
From an Enterprise point of view, ChatGPT can be used within bespoke or existing apps for customer engagement, report writing, content generation and more.
All conversational AI technology is powerful for customer service, and ChatGPT will likely be used in this area. And many more opportunities will surely arise as the technology grows. The more people play with it, the more we’ll discover what it can do.
Cautions and fears
Sometimes, ChatGPT is wrong. For example, we asked: “Explain how 2 to the power of 2 equals 8”.
2^2 = 2 * 2 = 4 * 2 = 8.
In other words, 2^2 represents the result of 2 being multiplied by itself 2 times, which equals 8.
Paul Sweeney, Chief Strategy Officer at Webio:
“The problem is the accuracy of the information. When you ask ChatGPT or other LLMs, they give you a very confident sounding response and you say, 'that seems right'. The bottom line, as John Oliver says, is 'truthiness'. How 'truthy' is it? It seems confident and the right answer, but we have no real sense of how true it actually is. The solution is to keep humans in the loop. It is a huge tool, but when the magic wears off, how do we actually make this work for us? The best solutions are still the well-engineered solutions, well structured, well tested.”
Fake news and perpetuating falsehoods are a real concern as the model grows.
Will it take over from humans in some areas? Since it is so good as generating content, this is one area that is making people feel a little uneasy, however, ChatGPT can have an assistant role in helping to write better and faster. Can we trust it to write news articles – what about journalistic ethics and critical thinking?
The ethical implications of ChatGPT are those of all AI technologies that cover areas like privacy, safety, justice and fairness. AI needs to be vigilant in keeping out bias, discrimination and aggression.
Copyright and ownership are other areas of concern. Already, students are presenting ChatGPT generated work as their own and academia will have a hard time controlling this.
Where to now for ChatGPT?
At the moment, ChatGPT is free to use (https://chat.openai.com/) and there is also a pro version that you can pay for that has better performance, gives you access at peak times and priority access to new releases.
In January 2023, Microsoft bought a $10 billion stake in OpenAI (it had already invested $1 billion in the early days of the company) and they have incorporated ChatGPT into its Bing search engine and some Office 365 apps (called Copilot).
In March 2023, OpenAI showcased the significantly more powerful GPT-4. Read more about what GPT-4 can do.
The way ahead seems bright for this technology. There is still a way to go, but there’s no turning back!