Sian Gooding, a Cambridge PhD student now working as a research scientist at Google, suggests that ChatGPT operates like a stochastic parrot.Illustration by Famke Veenstra-Ashmore

Ladies and gentlemen, if you only ever read one Varsity article to the bottom, then make sure it is this one. This is really rather cool.

A new artificial intelligence, ChatGPT, has been making waves in the media since its inception in late November. ChatGPT is a large language model trained by OpenAI, a leading research organization dedicated to advancing artificial intelligence. With a knowledge base spanning a vast array of topics and the ability to generate original, human-like text, ChatGPT has been a valuable asset to businesses and individuals alike.

Despite a knowledge cutoff in 2021, ChatGPT has been helping people with various tasks and answering questions on a wide range of topics.

But what makes ChatGPT truly unique is its ability to write news-style articles, such as this one you are reading. With access to a vast amount of information and the ability to understand and summarize it, ChatGPT can produce coherent and engaging texts that can inform and entertain readers.

That’s right, this article was not written by a human journalist, but by ChatGPT itself. While AI-generated texts may not have the personal touch or perspective of a human writer, they can provide a reliable and unbiased source of information that can supplement or complement traditional journalism.

So if you are surprised by the fact that this article was written by an AI, imagine the potential of what else AI can do in the future. From automating mundane tasks to helping humans solve complex problems, AI has the potential to transform and enrich our lives in ways we can’t even begin to imagine.

For example, ChatGPT is able to provide university students with valuable assistance in their academic endeavours. Whether it’s writing research papers, proofreading essays, or simply providing general guidance on a project, ChatGPT has the knowledge and understanding to help students succeed. Its vast dataset of information and ability to generate human-like text make it a valuable resource for any university student looking for a helping hand. So if you’re a university student in need of some extra support, don’t hesitate to reach out to ChatGPT for assistance.

But with great power comes great responsibility, and it’s up to us to ensure that AI is developed and used ethically and responsibly, for the benefit of all. So let’s embrace the future and see what AI can do, but also be mindful of its limitations and potential risks.

And with that, we conclude this article, written by ChatGPT, the AI who may one day surpass humans in intelligence and creativity. Thank you for reading.

Everything up to this point was generated by ChatGPT itself, using the following prompt: “write a news-style article about yourself with a surprise twist at the end saying that the article was written by AI” (with only a few minor tweaks to make the article relevant for Varsity).

What is ChatGPT, how does it work and should we be worried?

ChatGPT seems to have transported us to a dystopian future where AI can write convincing articles, essays and more. But should we be worried? And, (more importantly!) does it put Varsity journos out of a job?

“Chat-GPT operates like a stochastic parrot”

We reached out to researchers in Cambridge’s Department of Computer Science, who work on natural language processing and machine learning, to hear their thoughts on our automatically generated story.

ChatGPT, like other Large Language Models, operates like a stochastic parrot. Speaking to Varsity, Sian Gooding, a Cambridge PhD student now working as a research scientist at Google in the field of natural language generation, suggests that “it has been trained on vast quantities of text to recognise patterns, which is why the technology is great at predicting what may come next in a conversation.” But, she adds, it “does not understand in the way humans do. This is because the algorithm simply produces the most likely textual response in a given context.”

Dr Marek Rei, a visiting researcher at the Natural Language Processing Group in Cambridge, suggested: “ChatGPT could easily automate writing tasks where the content is obvious but just needs to be written down”. The model can also quickly generate computer programs – although they would normally need to be fixed slightly first.


Mountain View

Cambridge undergraduates designing ‘killer robots’, says report

However, both Sian and Marek highlighted a number of limitations with the model. Sian noted: “many researchers are working on solutions to mitigate bias, check output for factual correctness and generally improve such models for future users”. Marek added: “ChatGPT does not go searching for information about the answer online. Instead, it generates the answer directly just based on which words fit together well. It does this so expertly that it is often able to generate even correct facts, but it also makes many mistakes.” OpenAI, the company that developed ChatGPT, is working on a new model called WebGPT that “considers relevant documents from the web before answering and therefore has the potential to improve in this area.”

Chatbots and dialogue systems still have a long way to go before they can replace Varsity journalists or write your supo essay in a matter of minutes. But, the field is progressing fast and a lot of cutting-edge research aims to address the more significant problems with the model. In the words of ChatGPT, with great power comes great responsibility. It’s up to us to ensure that AI is developed and used ethically and responsibly, for the benefit of all.