The Nightly puts Elon Musk's article-writing AI to the test with The Matildas

Elon Musk’s X has soft-launched a new feature for premium users, an advanced chatbot curiously named Grok. Capable of replacing journalists, The Nightly put it to the test with Australia’s favourite sporting team.

Since the billionaire acquired the platform in 2022, there have been a number of teething issues and drastic changes.

Musk’s goal of slowly converting the app from a simple social media site to an all-encompassing hub for socialising, payment, video, and news has seen its latest advancement in the form of AI-generated headlines and stories using Musk’s “Grok” artificial intelligence service.

Those with access to X Premium will see a different version of the “Explore” tab on the sidebar, with robotic-sounding headlines and copy.

The body of the stories amalgamates several sources into one and has the distinct prose – and flaws – of artificial intelligence.

It has already caused issues, with Mashable reporting Grok had hallucinated an Iranian attack on the Israeli city of Tel Aviv and pushed that headline to potentially millions of users amid dense regional conflict.

Each story ends with the disclaimer “Grok is an early feature and can make mistakes. Verify its outputs,” before X compiles a series of top tweets about the subject.

Grok’s main advantage over ChatGPT is a much more dynamic information database, with X’s AI capable of “authoring” up-to-date sports reports using search results and posts from users on the app to build out the articles.

If you tweet about it, Grok can use it to form its article. The implications of such a system are obvious and dangerous.

So, if Grok struggles with determining fact from fiction when it comes to bloody conflict, how would it fare in a much less high-stakes story such as the Matildas friendly?

When The Nightly prompted Grok to create a match report for the Matildas match against Mexico, it could produce mostly competent copy, albeit with errors, stilted prose and clearly siphoned ideas from other outlets. Most notably, Katrina Gorry is both a star performer and also out injured.

Its struggle with names signals clear issues for the future, especially on stories of much higher significance — if it can’t determine if Katrina Gorry was playing or not playing in a friendly match, what chance would it have with political leaders saying or not saying things?

Grok’s match report, while serviceable, doesn’t feel like it’s actually saying anything from a point of knowledge. It can’t do added insight, it’s simply a blend of numbers out of the match statistics and snatched elements of human reports. Which is the uncomfortable reality of Grok (and all AI).

Most people probably don’t know their tweets are being used for this at all, and that doesn’t feel right.

Grok is an astonishing tool if used correctly, but how it works doesn’t pass the pub test. Knowing that every tweet feeds the AI also sets it up for human interference - all you’d need is some bots and an agenda.

It’s amazing it exists, it’s fun to play around with, but it can’t really replace journalists or writers — and we can only hope it doesn’t.