Testing Out ChatGPT

On December 15, OpenAI updated its recently-released, transformative AI platform, ChatGPT, a Generative Pre-training Transformer (GPT) that, according to its self-description, can generate "high-quality text that is difficult to distinguish from text written by humans." More advanced than typical chatbots, this type of AI can create new content, such as text, images, or music, based on inputs or information OpenAI has fed it from the internet and through textual conversations to best mimic human dialogue. ChatGPT can write passages, create reports, draft key messages, formulate code, and design graphics in seconds. And it's free, for now.

In line with what we do at the Institute for Public Relations (IPR), I asked ChatGPT to "explain psychological safety in simple terms" to compare its output with a recent piece we did at IPR. Here's the ChatGPT output:

This is a much shorter version than the IPR Deconstructing: Psychological Safety Research Brief that breaks down commonly misunderstood constructs and terms used in public relations and communication. In a matter of seconds, ChatGPT produced an accurate and easily digestible brief of what psychological safety is. I spent a few hours on ChatGPT and while it is extremely useful in some ways, it's limiting in others. Here's what I found:

• ChatGPT doesn't always cite sources. Currently, the program is like a poorly trained undergraduate student who doesn't cite their sources. Plus, it's not always accurate. As the program explains, "ChatGPT is not connected to the internet, and it can occasionally produce incorrect answers."
• ChatGPT may not be updated on current sources. Because ChatGPT has an established information database (and doesn't scour the internet to generate a response), the program has limited information post-2021. It admits that it "may also occasionally produce harmful instructions or biased content."
• ChatGPT's response is fairly general in terms of application. The recommendations we provided in our IPR piece were more research-based, relevant, and specific. ChatGPT's recommendations were general or as we like to describe these types of results at IPR, fairly "Captain Obvious."

Overall, the IPR version is far superior to the ChatGPT version, but ChatGPT is an excellent source and primer for many topics. One thing IPR can't do that ChatGPT can do is to shake things up a bit when it comes to writing the results. In a matter of seconds, ChatGPT cranked out research poetry when I asked it to "explain psychological safety in the style of Shakespeare." Here's a stanza:

Now that is amazing! ChatGPT can construct phrases or stories in the style of artists like Eminem and Beyonce, but when I asked ChatGPT to construct a bedtime story in the style of my favorite bands, Death Cab for Cutie or Wilco, the stories were fairly indistinguishable, inserting the name of the band in places rather than constructing a story in the style of the music. Maybe ChatGPT isn't a fan of alternative rock or indie music....yet.

I know there may are concerns about the ethics of this type of AI and the power of generative technology. ChatGPT has been primed by OpenAI, its parent company, to filter out content that may contain inappropriate, harmful, or unethical requests (i.e., racist or biased). While there are some checks in place, some requests slip through the cracks.

ChatGPT does rightly avoid some value-based questions. For example, I asked ChatGPT why IPR is the best organization and it replied, "It is not appropriate for me to make a statement about the superiority of any particular organization." ChatGPT has also experienced times when it is "too busy" because of the high number of visitors on the site. As a generative technology, though, ChatGPT has the ability to learn from its mistakes. As ChatGPT matures, this AI could greatly shift how we work and focus our time.

Big thanks to Jeremy Gilbert at Northwestern University for sharing his knowledge and passion about ChatGPT and other technologies in the IPR Master Class last week. To learn more about ChatGPT and its application, check out The Daily podcast episode from Friday, Dec 16, "Did Artificial Intelligence Just Get Too Smart?" It's worth a listen. And if you want to try it for yourself, visit the ChatGPT site.

I would love to hear from you. How are you using ChatGPT? What do you think about it? How will that affect how we work?