A new conversational computer model called ‘ChatGPT’, released last month by the company OpenAI, has sparked global curiosity.
It has led to predictions of the end of Google, though OpenAI’s own CEO calls it “extremely limited” and says it would be a “mistake” to rely on it.
What is it, exactly?
What is ChatGPT?
ChatGPT is a “large language model”. Its creators have “trained” it by making it read vast amounts of text from the internet. It uses what it has read to answer user questions in a conversational style.
It’s capable of producing a wide range of texts including essays, jokes, recipes, movie scripts, code, and emails.
How good is it?
Some users have shared promising results. One university professor asked the bot to write an essay with a result that was coherent and original enough to pass through anti-plagiarism software.
Positive results have also been reported in ‘smart’ search functions that would be difficult to achieve on Google, like explaining the humour in a joke.
In other ways, it remains limited – by its own admission (we asked it), it doesn’t “understand” the text it generates, which can lead to nonsense answers and obvious factual errors.
OpenAI CEO Sam Altman has spoken bluntly about its current shortcomings.
“ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness,” Altman tweeted recently. “It’s a mistake to be relying on it for anything important right now. It’s a preview of progress; we have lots of work to do on robustness and truthfulness.”
The popularity of the tool despite its limitations has led to debate among experts. Paul Kedrosky, a venture capitalist and researcher at the Massachusetts Institute of Technology, called it a “nuclear bomb” at a time when society is “unprepared” to appreciate its limitations, and called for it to be withdrawn “immediately”, in a since-deleted tweet.
There has also been debate about how much its findings should be restricted. There have been examples of models trained on internet text reproducing prejudices that sit behind those texts.
Can it do journalism?
ChatGPT can’t search the internet and isn’t aware of current events or news, which means The Daily Aus is safe for now.
When asked to write a short summary about Prime Minister Anthony Albanese, ChatGPT politely responded that it did not know him. “If you have specific information about Anthony Albanese that you would like me to use, I would be happy to try to help.”