If you have not heard of ChatGPT, where have you been since November 30th when it was launched by OpenAI the company that developed what is fast becoming a groundbreaking technology. Since then, it’s been making waves, everyone is talking about and using it. In the first week alone over 1,000,000 people had subscribed to what is for the time being at least, free. OpenAI was founded in 2016 by Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba, and Sam Altman, although Musk stepped down from the board in February 2018 to avoid any conflict with Tesla.

Originally a “not for profit”, in 2019 it became a “capped for-profit”, meaning in this instance that investors can’t earn more than 100 times their investment. And just to give some idea of scale, OpenAI expects to generate $200 million in revenue next year and $1 billion by 2024.

Its mission is simple, yet as you might imagine ambitious – to ensure that artificial general intelligence benefits all of humanity.

In contrast here is Googles mission – to organise the world’s information and make it universally accessible and useful.

But what is ChatGPT? – To find an answer to this question, we can actually ask ChatGPT, however before we do there are some points worth noting. ChatGPT is not a search engine unlike Google, it’s not even connected to the internet, in fact the data sets used to train it haven’t been updated since the end of 2021. You will find this out if you ask anything relatively topical, for example, who won the World Cup this year, it will say….

“I’m sorry, but I am not able to provide information about events that have occurred after my knowledge cut-off in 2021”.

The answers appear in real time in a suitably bot like fashion, but what makes it really interesting is that it uses Generative AI, which enables it to start with existing content but from that produce something new and original, let me clarify, its not simply taking content from somewhere else, its creating new and original stuff.

ChatGPT is a variant of the GPT (Generative Pre-trained Transformer) language model developed by OpenAI for use in chatbots and conversational systems. It is able to generate human-like text in real-time and understand the context of a conversation, allowing it to generate appropriate responses based on the topic being discussed. ChatGPT is also efficient and fast, making it ideal for use in chatbot applications where fast response times are important. It is able to handle a wide variety of tasks, including answering questions, providing information, and engaging in casual conversation.

The above was the answer provided by ChatGPT when asked for a definition within 90 words. In fact everything in purple and italics in this blog has been taken directly from ChatGPT.

How does it learn?
Once again let’s ask ChatGPT, after all why should I do all the hard work.

ChatGPT uses a type of machine learning called supervised learning. In supervised learning, the model is trained on a labeled dataset, where the correct output is provided for each input. The model is then able to make predictions for new, unseen data based on the patterns it learned during training. too train ChatGPT, a large dataset of text is fed into the model, and the model is then trained to predict the next word in a sequence based on the words that come before it. Through this process, the model learns to recognize patterns in language and generate coherent text. The more data the model is trained on, the better it becomes at generating natural-sounding text and understanding the relationships between words and concepts

In summary, the model is initially trained by humans (The labeller), this is the supervised stage. It is then able to provide multiple answers to a question of which the trainer ranks in terms of quality, this is the reinforcement stage which provides more data for the model. And lastly, reward models are created that automatically reward the model for getting it right, the unsupervised stage.

It might be easier to think of it in the same way as you would if learning to ride a bike, for example firstly someone shows you how to ride a bike, (supervised) secondly you have a go and are given feedback in order to improve (reinforcement) and finally, if you ride for 5 minutes, you will be rewarded with £5.00, fall off and you get nothing. (Reward model – unsupervised).

Clever… but what are the implications?
We are at one of those “genie out of the bottle” moments, when something that you thought might happen in the future becomes a reality today. As a consequence, we start to ask questions such as, is this technology good or bad, what will it mean for jobs and the future of work? If it can produce high quality answers to questions, how can we tell if it’s the student’s work or simply the result of an exercise in cut and paste? And because it can write poems, stories and news articles, how can you know if anything is truly original, think deep fake but using words. By way of an example, here is a limerick I didn’t write about accountants.

There once was an accountant named Sue
Who loved numbers, they were her clue
She worked with great care
To balance the ledger with great flair
And made sure all the finances were true

Okay it might need a bit of work but hopefully you can see it has potential.

We have however seen this all before when other innovative technologies first appeared, for example, the motor car, the development of computers and more recently mobile phones and the internet. The truth is they did change how we worked and resulted in people losing their jobs, the same is almost certainly going to be the case with ChatGPT. One thing is for sure, you can’t put the genie back in the bottle.

Technology is neither good nor bad; nor is it neutral. Melvin Kranzberg’s first law of technology

And for learning
There have already been some suggesting that examinations should no longer be allowed to be sat remotely and that Universities should stop using essays and dissertations to asses performance.

However, ChatGPT is not Deep thought from The Hitchhikers Guide to the Galaxy nor Hal from 2001 a Space Odyssey, it has many limitations. The answers are not always correct, the quality of the answer is dependent on the quality of the question and as we have already seen, 2022 doesn’t exist at the moment.

There are also some really interesting ways in which it could be used to help students.

• Use it as a “critical friend”, paste your answer into ChatGPT and ask for ways it might be improved, for example in terms of grammar and or structure.
• Similar to the internet, if you have writers block just post a question and see what comes back.
• Ask it to generate a number of test questions on a specific subject.
• Have a conversation with it, ask it to explain something you don’t understand.

Clearly it should not be used by a student to pass off an answer as their own, that’s called cheating but it’s a tool and one that has a lot of potential if used properly by both students and teachers.

Once upon a time, sound was new technology. Peter Jackson filmmaker

PS – if you are more interested in pictures than words check out DALL·E 2, which allows anyone to create images by writing a text description. This has also been built by OpenAI.