I tried OpenAI's ChatGPT 2 years ago. I'm terrified by how far it's come and think it could change my job.

• I'm a writer who used GPT-3, an early version of OpenAI's viral chatbot, in 2021.
• When ChatGPT was released to the public, I checked in to see how it had evolved.

Two years ago, I was a ghostwriter for a company that was licensed to use what was then called GPT-3, developed by OpenAI.

The latest artificial-intelligence iteration released from OpenAI is known as ChatGPT. The hype around its ability to produce human-like written content is raising questions about which professional roles it could make redundant — and worrying writers, like me.

Having had the pleasure of getting to know the bot before it was famous, I compared the version I used two years ago with the version that became publicly available in November.

Complimentary Tech Event

Transform talent with learning that works

Capability development is critical for businesses who want to push the envelope of innovation.Discover how business leaders are strategizing around building talent capabilities and empowering employee transformation.Know More

This experience has given me some insight into what ChatGPT could mean for the future of the writing profession.

I first encountered GPT while working as a writer for a company that used AI to improve enterprise chatbots through machine learning

I was working for a company that was basically creating chatbots that could have less-rigid conversations, with the possibility they could become "personal assistants" for businesses.

This company was licensed to use GPT-3 in 2021 because OpenAI wanted to test its model with stakeholders before introducing it to the public.

I ghostwrote an article we pitched to TechRadar titled "Chatbots and GPT-3: Using human knowledge and relevant context for better chatbot experiences."

The reader had to guess which paragraph was written by the AI. We'd run a final draft through GPT and asked it to write a paragraph similar to my prose, with a disclaimer at the bottom saying that AI had written one of the paragraphs.

After I gave the AI my writing, one of the company's AI experts "primed" the GPT for our topic and provided it with some of my past writing so it could mirror my style. It spit out two results, and I picked which sounded most like me.

My initial reaction was that this machine was pretty darn good, but the writing had some giveaways

I was impressed, but the formatting was pretty machinelike. The first example was one big paragraph. In the second, all the paragraphs were two sentences long. This was almost too broken up and regimented.

If it were me, I would've broken it up based on the ideas to make it more intuitive and give it the narrative rhythm.

All things considered, the GPT was able to write in my style and tone and grasped the concept of the article as a whole.

What was more impressive, albeit a little scary, was the AI deployed conversational phrases that I often used in my writing, such as "aha moment" and "dead air," which made the sample seem more human.

I chose the second option, and the article was published.

When ChatGPT was released in November, I wanted to see how far my old friend had come

I registered with OpenAI and started to play around.

I fed the current version of ChatGPT my original article draft that had been through the AI two years ago to compare the AIs' capabilities.

I first instructed the bot to "write a paragraph based on this draft." It pretty much just regurgitated the whole piece in a five-sentence overview. On closer inspection, the AI had actually lifted sentences from the original article and put them back together.

Being in the driver's seat this time, I realized I needed to give the AI more instruction.

I asked the chatbot the same question as before but added a request for a "new paragraph" using a technical style with a conversational tone. It still summarized the draft into one paragraph but didn't plagiarize, and it drew some conclusions based on the content that wasn't written in the original piece.

Still, I wanted something other than a summary, so the third time I got very specific.

I said: "Write a new paragraph based on the draft I provided, but write it not as a general summary of the content. Rather, write the paragraph so that I can put it into one part of the draft seamlessly. Use a technical style with a conversational tone, and do not format it as a large paragraph."

This time, the output showed me just how far this AI had come. I was terrified because all of my fears from when I first encountered it were coming to fruition, yet the failsafe limitations were no longer there – it was much, much smarter now.

It was able to draw conclusions from the draft and write, dare I say, original ideas based on it, also using tactics for writing well, such as varying the length of the sentences.

It even used some humanlike phrases to describe things, and most impressively, it was able to intuitively explain the limitations of its own abilities.

The key lesson from my experiment was summarized best by the bot itself. ChatGPT is a "tool for language-based solutions rather than an intelligent conversationalist," it wrote after my third request.

I think ChatGPT is a game changer for writers and content creators. These were my 3 takeaways from my experience.

1. Writers' roles won't disappear with the introduction of AI but will evolve more toward communication strategy. Writers and editors will be directing the AI's writing. Successfully employing the bot comes down to giving it clear instructions. The AI still needs to be supplied with specific goals, a target audience, and a tone of voice. It also needs to be asked the right questions.
2. It has to be primed for a topic and given past content so it can imitate writing accurately. If writers want ChatGPT to produce worthwhile work, they will need to give it large amounts of contextual data. This will require writers to do thorough research and fact-checking because the AI could very well pull from the millions of inaccurate sources on the internet if not guided correctly.
3. When the internet becomes flooded with half-hearted AI writing, people working in editorial roles will need to set their writing apart with human touches. Chatbots will become increasingly better at generating ideas and sampling text, which will make point-of-view reporting even more critical — perhaps through a personal perspective on a story or first-person interviews.

I think any future version of ChatGPT will need a human to fine-tune the final product. This will require writers and editors to develop their skills toward understanding how to implement and manage AI. So the sooner they can embrace ChatGPT, the better.