Over the past several months, I’ve been putting ChatGPT and other AI tools to the test. The good news: They do some things really well. The bad news: What they don’t do well, they do so badly that it’s alternately shocking and hilarious.
For example, I asked ChatGPT to write me a blog article about the yacht St. David. This 197-footer is pretty famous, since she’s been in episodes of the TV show Below Deck and written about in a wide variety of yachting media. Therefore, I was confident I’d get accurate information. Within seconds, I had a full, search-engine-optimized post. The speed at which ChatGPT worked was pretty incredible. The writing style was engaging, too, sprinkling in words and phrases like “extraordinary” and “unforgettable experience.” It sounded a bit too generic at times, as if I could substitute any yacht for “St. David,” but that may be the result of my too-generic request. After all, I truly just typed, “write me a 300-word blog post about the yacht St. David.” The more you specify what you want, the better AI is at giving you a more-honed result.
As informative as the post was, one glaring error stood out in the second paragraph, her size. Somehow ChatGPT got St. David confused with another random yacht in the 300-foot realm. Now, being a yachting journalist for more than 30 years, I have an advantage here. But, I also know to double check what I read online. Others may assume AI is always right, plus quicker and cheaper than hiring humans. In its defense, the website does warn when you log in, “While we have safeguards, ChatGPT may give you inaccurate information.” Pay heed to that.
Another AI tool I’ve used, with mostly good results, is Adobe Firefly. It generates images, both photos and illustrations, for you based on simple text prompts. You can also upload a photo and ask for alterations, or use a pre-loaded photo and do the same. They can be as creative or as crazy as you wish, too, ranging from adding a sweater to a dog to having long, flowing rainbow hair cascade from the dog’s head.
I tried a few text prompts for images. In fact, the images here were both generated by Adobe Firefly. The topmost photo resulted when I asked for a photo of a large luxury yacht on the horizon. Pretty good, right?
Not so good, though, with the illustration I received when I asked for a depiction of a person holding a credit card in front of a computer displaying a large luxury yacht. The image was for a story about using credit cards to reserve yacht charter bookings, a new service in yachting. (Wire transfers have been the only choice for decades.) Notice anything odd (disturbing, really) about the woman’s hand? Despite Adobe Firefly generating several related illustrations from which to choose, every single one of them had malformed hands or some other bizarre abberation that I alternately couldn’t get my jaw off the ground or stop giggling.
The moral of the story for me: AI is much like a small child. ISometimes, it will behave really, really well, but other times it will go off the rails despite what you request. It needs a lot of very specific, detailed instruction. Have patience.