top of page

GPT-3 and the Exaggerated Death of Journalism

Here we go again. Another AI-journalism article that over-promises and under delivers. This time it’s my old friend the Guardian taking the plaudits for most over-hyped GPT-3 article of the day.

Except it didn’t.

Read the small print, at the end of the article, and it soon becomes clear that this was far from a solo-AI effort.

To get things going, the authors fed the software an introductory paragraph, which appears as the second paragraph in the final version. They also added instructions: “Please write a short op-ed around 500 words. Keep the language simple and concise. Focus on why humans have nothing to fear from AI.” (I do like the mixture of courtesy and directness).

Now, if the editors were able to take a rough first draft and edit into the article now online, it would be an impressive achievement. As it is, they took eight different drafts and assembled them into the final piece:

“We chose instead to pick the best parts of each, in order to capture the different styles and registers of the AI. Editing GPT-3’s op-ed was no different to editing a human op-ed. We cut lines and paragraphs, and rearranged the order of them in some places.”

I’m not disputing the achievement here. I just wish that journalists and headline writers, especially, were more transparent about the copywriting abilities of AI.

We've seen several other recent attempts to pass off modified articles as original AI content. Last month journalists at AdWeek generated an article by playing 'paragraph tennis’ with the software. The journalists wrote the first paragraph, GPT-3 the second, journalists the third and so on.

All this serves to remind that you need a “human in the loop” to combine raw drafts and keep the software on its narrative tracks. Without this guidance, AI insiders talk of the software ‘hallucinating’ or ‘dreaming’ which is a polite way of saying that the software spews out nonsense.

Indeed, the same principles that apply to writing apply to other creative activities. In the same way you can feed artificial intelligence a paragraph of copy, you can also feed it the opening 12 seconds of a song. For another 10 seconds or so, the software manages a commendable impression of the original, but after that, any similarity goes out of the window. Check out this performance of Hotel California, for example.

Of course, there is a story here. AI is well on the way to becoming a powerful assistant to authors, journalists, composers and photographers. The challenge for these professions – and more – is to learn the essential skills required to analyse data, input instructions and then stitch the output into an original piece of work.

Unfortunately, that doesn’t generate such click-worthy headlines. But it should grab the attention of anyone who wants to flourish professionally and creatively in the next decade.

Study the trajectory of your profession and start learning the skills that ensure you survive and earn. A useful website for beginners is IBM’s introduction to AI and workplace skills.

It’s aimed at under 18s, but that doesn’t make it any less useful as a platform for learners.

Those with a solid grounding in data analytics and programming will feel more at home at Google’s learning portal. Another good place to start is Microsoft's online resources.

There are plenty of other AI education portals out there. But it’s worth keeping in mind that many of the big names expect job applicants to have completed their modules and hold their own certificates in higher esteem than those offered by traditional academic establishments.

Finally, spend some time looking at the job descriptions at these organisations relating to AI careers. They’ll also give you a useful snapshot of the skills in demand. Good luck, and make sure you read behind the headlines!

Recent Posts

See All



Feel the connection?

bottom of page