ChatGPT’s writing capabilities are staggering, but humans are still essential (for now)

If you’ve spent any time browsing social media feeds in the past week (who hasn’t), you’ve probably heard of ChatGPT. The bewitching and mind-blowing chatbotdeveloped by OpenAI and released last week, is a nifty little AI that can spit out very compelling, human-sounding text in response to user-generated prompts.

You could, for example, ask him to write a plot summary for Knives Out, except that Benoit Blanc is actually Foghorn Leghorn (just me?), and he’ll spit out something relatively coherent. It can also help fix broken code and write essays so compelling that some academics say they’d get an A on college exams.

His answers stunned people to such a degree that some even proclaimed, “Google is dead.” Then there are those who think it goes beyond Google: human jobs are also in trouble.

The Guardian, for example, proclaimed that “teachers, programmers and journalists could all be out of work in just a few years.” Another take, from the Australian Computer Society’s flagship publication, Information Age, suggested the same. The Telegraph reported that the bot could “do your job better than you”.

I would say hold your digital horses. ChatGPT is not going to put you out of work just yet.

A prime example of why is provided by the story published in Information Age. The publication used ChatGPT to write an entire ChatGPT story and published the finished product with a short introduction. The article is about as simple as you can ask for – ChatGPT provides a basic account of the facts of its existence – but in “writing” the article, ChatGPT also generated fake quotes and attributed them to a researcher. ‘OpenAI, John Smith (which is real, Apparently).

This underlines the main flaw of a large language model like ChatGPT: it doesn’t know how to separate fact from fiction. He can’t be trained to do that. It’s a word organizer, an AI programmed in such a way that it can write coherent sentences.

That’s an important distinction, and it essentially prevents ChatGPT (or the big underlying language model it’s built on, OpenAI’s GPT 3.5) from writing news or talking about current affairs (it doesn’t is also not formed on up-to-date data, but that is something else). He certainly can’t do the job of a journalist. Saying that diminishes the act of journalism itself.

ChatGPT will not go out into the world to tell Ukrainians about the Russian invasion. He won’t be able to read the emotion on Kylian Mbappe’s face when he wins the World Cup. This is definitely not jumping on a boat to Antarctica to write about his experiences. He can’t be surprised by a completely irrelevant quote that unknowingly reveals a CEO’s business secret. Hell, he would have no hope of covering up Musk’s takeover of Twitter – he’s not an arbiter of truth, and he just can’t read the play.

It’s interesting to see how positive the response to ChatGPT has been. It’s absolutely commendable, and the documented improvements that OpenAI has made to its latest product, GPT-3, are interesting in themselves. But the main reason it really grabs attention is because it’s so easily accessible.

GPT-3 lacked a simple, easy-to-use online framework, and although publications like the Guardian used it to generate stories, it only had a brief splash online. Developing a chatbot that you can interact with and share screenshots completely changes the way the product is used and talked about. This also contributed to the bot being a bit overhyped.

Curiously, it’s the second AI to make a splash in recent weeks.

November 15, Meta AI has released its own artificial intelligence, dubbed Galactica. Like ChatGPT, it’s a great language model and has been touted as a way to “organize science”. Essentially, it could generate answers to questions like “What is quantum gravity?” or explain mathematical equations. Just like ChatGPT, you ask a question and it provides an answer.

Galactica has been trained on over 48 million scientific papers and abstracts, and it has provided compelling answers. The development team pitched the bot as a way to organize knowledge, noting that it could generate Wikipedia articles and scientific papers.

The problem was that it was mostly garbage pumping – nonsense text that seemed official and even included references to scientific literature, although these were made up. The sheer volume of misinformation he produced in response to mere prompts, and how insidious that misinformation was, bugged academics and artificial intelligence researchers, who let their thoughts fly on Twitter. The backlash saw the project shut down by the Meta AI team after two days.

ChatGPT doesn’t seem to be going in the same direction. It looks like a “smarter” version of Galactica, with a much more powerful filter. Where Galactica offered ways to build a bombshell, for example, ChatGPT eliminates discriminatory, offensive, or inappropriate requests. ChatGPT was also trained to be conversational and admit mistakes.

And yet, ChatGPT is still limited in the same way as all major language models. Its purpose is to construct sentences or songs or paragraphs or essays by studying billions (trillions?) of words that exist on the web. It then puts those words together, predicting the best way to configure them.

In doing so, he writes some pretty compelling essay answers, of course. It also writes garbage, just like Galactica. How can you learn from an AI that might not provide a truthful answer? What kind of jobs could it replace? Will the public know who or what wrote a play? And how can you know the AI ​​isn’t truthful, especially if it sounds convincing? The OpenAI team recognizes the shortcomings of the bot, but these are unresolved issues that limit the capabilities of an AI like this today.

So, even if the little chatbot is entertaining, as evidenced by this wonderful exchange about a guy who brags about pumpkinsit’s hard to see how this AI would put professors, programmers or journalists out of work. Instead, in the short term, ChatGPT and its underlying model will likely supplement what journalists, professors, and programmers are doing. It is a tool, not a replacement. Just as journalists use AI to transcribe long interviews, they can use ChatGPT-like AI to, say, generate a headline idea.

Because that’s exactly what we did with this piece. The title you see on this article was, in part, suggested by ChatGPT. But his suggestions were not perfect. He suggested using terms such as “Human Employment” and “Humans Workers”. These seemed too official, too… robotic. Void of emotion. So we tweaked his suggestions until we got what you see above.

Does this mean that a future iteration of ChatGPT or its underlying AI model (which could be released as early as next year) won’t show up and render us useless?

Maybe! For now, I feel like my job as a journalist is pretty safe.