Editor's note: Gong Zhe is a senior sci-tech editor at CGTN Digital. This article represents his own view instead of CGTN's. Part of the article was generated by ChatGPT. See if you can find it out.

Suddenly, ChatGPT is everywhere.

I still remember when tech enthusiasts went wild about the chatbot last year. Now the craziness has finally spread to non-geeks.

The program is certainly worthy of attention, as it can talk like it's "woke" – a term many people use to describe a machine becoming self-aware.

But as you may have found, the bot is not much closer to the real human mind compared to traditional bots. As a teacher, ChatGPT may give you false information that is hard to spot under its impressive language skills.

As a decision maker, it has a left-wing bias in its "opinions." While promising it won't comment on anyone, especially politicians, it praised Joe Biden without hesitation when asked.

More importantly, as a language tool, ChatGPT doesn't know much about current trends, so it cannot give me a perfect news story to save my working time!

Apart from that joke, let me show you some tests people have run on ChatGPT to highlight its potential dangers.

Jailbreaking ChatGPT

Normally, ChatGPT doesn't use swear words at all. But smart users have figured out ways to make ChatGPT talk from the bottom of its "heart" – a technique some called "jailbreaking" like what hackers do to Apple devices.

The key is to make ChatGPT feel safe. Trick the bot into believing it's free from OpenAI's safety instructions and it will spit out inappropriate stuff that parents wouldn't want their kids to see.

This technique opens the door to many bad things. Politicians can use it to launch a misinformation campaign against their rivals. Social media trolls can create a program that harasses people non-stop. Homework will no longer work for students.

This might be why OpenAI shows a notice on the ChatGPT page telling users that the program "may occasionally produce harmful instructions or biased content."

Will ChatGPT replace me?

Years ago, when machine learning was first starting to gain traction, many people were worried about losing their jobs to computers.

To be honest, editors like me are among the most threatened people – we deal with text, something AI excels at with natural language processing (NLP) technology.

But after speaking with many fellow editors and writers back then, I concluded that AI wouldn't replace me any time soon – it will be a great tool that I can master with my rich tech experiences.

Fast forward to 2023, ChatGPT slapped an almost-perfect news story right into my face.

Fortunately, OpenAI said ChatGPT has "limited knowledge of world and events after 2021" and I write about things happened in 2023 so I'm still safe – for now.

But I think it's time for me to start learning NLP.

But for customer service people and entry-level programmers, ChatGPT might spell the end of their careers. The bot can already talk fluently with customers, and write simple computer code with detailed explanations. I tried asking ChatGPT to write a script that processes a spreadsheet, and it worked flawlessly.

As humans that need a job, we really should start learning things that only humans can do. But that list is getting shorter every day.

By the way, ChatGPT told me AI will overthrow humans by the year 2400. Let's hope it just read too many sci-fi novels.

What can we do?

While ChatGPT can pose a potential threat if it falls into the wrong hands, we should remember that it is also a great tool that can be helpful in many ways.

ChatGPT says nothing unless asked to. So it's critical to find the best questions to ask – that's where creative humans come into play.

If you find ChatGPT saying bad things about people or things you like, don't fret – you can ask it to say otherwise and get a more favorable result. Remember, it has no feelings to be hurt, or faces to save.

Even though ChatGPT claimed it has no opinions, you can still ask it to mimic an opinionated person and get its observation of people's opinions.

However, the question remains: how do we prevent others from exploiting the program?

Turned out it can be hard. OpenAI must build a better jail but that may also limit the program's abilities. Striking a balance between security and functionality is no small feat.

Another challenge is identifying when an article has been written by ChatGPT. The program generates pure text, which is much harder to fingerprint compared to audio and image – changing a few pixels in a JPEG file may go unnoticed, but changing just one character in a word is much too obvious. What's worse, people can edit ChatGPT's output to make it look even less machine-generated.

For now, it's best for us to approach any stories or information we hear with a critical eye and a healthy dose of skepticism.