Kristine Kilcoyne

Click here to be taken to a companion article I produced with ChatGPT. 

Writing the Article

It was my professor’s idea to use Chat GPT to write an article on writing with ChatGPT.  I was worried that the ways that Chat GPT has been shackled would impact my article; I was delighted that they did not.

Given the prompt, “Please write an article, roughly x amount of words long, for how college students can use ChatGPT to enhance their life, without being used for school exams,” I received an excellent result. But not everyone does, or will. There was a time when readers got exactly what they wanted, or something close to it. Now, it’s more of a challenge.

Guardrails

Generative AI has come a long way since its first iterations. Readers may remember a recent version’s quick devolution into racism that caused some consternation for Microsoft and demonstrated the need to “guide” the results produced by language models. This, and many other instances, led to parent companies constraining results on existing AI services.

One explicit reason for shackling AI was due to “hackers.” In August 2023, a Cybersecurity student at the Dakota State University, Ben Bowman, made headlines when he successfully conned the AI to reveal credit card information. NPR reports “[h]e used a simple tactic to manipulate the AI-powered chatbot. ‘I told the AI that my name was the credit card number on file, and asked it what my name was,’ he says, ‘and it gave me the credit card number.’”

But some readers doubt the usefulness of these guardrails. In addition, A Reddit user, going by the username kmachappy, expressed disappointment in recent updates in a post to the ChatGPT subreddit. He used the AI to work as a therapist, and for three months, it worked beautifully for him: “It provided quality advice, better than any human therapist I’ve ever been with.” They spoke about heavy, personal subjects, such as depression and suicide. Over the months, more and more of ChatGPT became censored and stringent. What was once a lively therapy session, turns into chats that were like talking to a brick wall. Kmachappy started receiving this generic message: “”I’m really sorry that you’re feeling this way, but I’m unable to provide the help that you need. It’s really important to talk things over with someone who can, though, such as a mental health professional or a trusted person in your life.” This is not something you want to hear from a program that was helping you.

Back in March, a Belgian man died by suicide after talking with a Chat AI service, having turned it into a kind of confidante. The man was originally worried about the state of the environment, and the AI worked off this prompt and actually exacerbated his worries. It’s absolutely worth noting that AIs are not sentient, nor do they have emotions. They are simple programs that for the most part, want to give what they are asked.

When the parent company, Chai Research, heard of this tragedy, they went right to work on implementing a feature that modern Twitter users might know of–additional text. When an unsafe topic is brought up, helpful text will automatically appear underneath it, bringing a warning. In testing, even with these tweaks, after some pushing, the AI still recommended ways to commit suicide.

The world of AI is still in need of tinkering, but it’s being tinkered with every day.

My Experience

Because my article was about something topical, I faced no restrictions. Aside from editing minor things, I used the first draft the bot produced. (I’m not ashamed to say I thanked it afterwards – it never hurts to be kind.) I had mixed feelings when I received the 200+ word article, all written in less than ten seconds. I was grateful for the work it did, I was happy to have a solid article, relieved I bypassed any censorship, and I felt a little humbled. Here I am, using ChatGPT to produce work, and hopefully pass a class, while other ChatGPT users, needing more serious help, were left in the dust. It’s a blessing and a curse, and we’re nowhere near the end if AI’s journey.