š¶ On this fourth day of Christmas, StrefaTECH explores ā¦ Four AI Calling Cards š¶
With the release of ChatGPT in late 2022 and humanity being what it is, some of the earliest explorations of the new tool were for objectives ranging from lazy (e.g., students cheating on homework) to nefarious (e.g., ransom calls with AI-generated fake voices). And in the never-ending game of whack-a-mole, software developers and well-meaning entrepreneurs have been seeking ways to catch, if not prevent, those uses. There have been numerous stories on the theme of how to detect AI-generated content.
Here we are a year later, and sadly, no dependable system exists to detect AI.
But! We humans still possess certain capabilities not yet fully realized in computer software (ahem, yet), such as our adeptness at sniffing out fake content. Weāre not flawless, weāre sometimes gullible, but we feel really good when we ferret out some secret!
So for todayās Friday Fun, Iām hoping youāll enjoy my somewhat whimsical ā4 AI Calling Cardsā tome describing a few telltale signs that what youāre reading or gazing at probably had some AI chatbot involved in its creation.
Calling Card #1: Emoji Overload
I prompted ChatGPT to help me out here, asking it to: Write an article describing and illustrating how ChatGPT-generated content often is structured as a series of sentences or short paragraphs, each starting with a different colorful and creative emoji.
Its response was spot-on!
In today's exploration, we examine how ChatGPT uses emojis to structure its conversations.
š Friendly Beginnings: ChatGPT often starts with a smiley face emoji, setting a welcoming tone. This is like a digital handshake, especially useful in introductory messages in the nonprofit tech world.
š” Insightful Ideas: The light bulb emoji signals the introduction of new concepts or solutions, like when discussing innovative tech strategies for nonprofits.
š Knowledge Sharing: The book emoji denotes in-depth information, perfect for detailed discussions about software or digital tools in nonprofit settings.
š Detailed Analysis: The magnifying glass emoji cues analytical content, ideal for examining trends or challenges in technology affecting nonprofits.
š Global Context: Finally, the globe emoji introduces broader perspectives, reflecting the worldwide impact of technology on the nonprofit sector.
When I first was using ChatGPT to draft content for emails, event pages, etc., I was somewhat charmed by the emojis. Now that Iāve both published and received a few dozen items formatted in that theme, though, Iāve become jaded and even sometimes a bit annoyed.
By the way, if you ask other chatbots (e.g., Claude, Bard, Pi) a similar question, theyāll explain how they use emojis to enhance communication, too!
š«¢My AI-sniffer turns its nose up at emoji overload. š«¢
Calling Card #2: Peppy, Peppier, Peppiest
In the 70s, there were bands in the soft pop/rock genre whose music was peppy. Lyrics were happy, melodies were singable, harmonies were easy. Think Abba, Air Supply, The Bee Gees.
When I read drafts written by many of the chatbots, I find myself humming āFernandoā (and wish I hadnāt pointed that out, as now Iām going to have trouble getting it out of my head!). In my dialogue just now with Claude-2, where I asked some questions about those bands and their signature songs, it gave a wonderful synopsis: The heavy orchestration and emotional vocals on these songs dripping with sentimentality make them almost sickly-sweet examples of that soft pop/rock era!
Iāve tried using custom instructions in ChatGPT to tone down the peppy-meter. After a few rounds of adjustments and with some help from ChatGPT to frame these, hereās where mine have settled (the second box in custom instructions, How would you like ChatGPT to respond?):
Tone: Opt for a professional yet accessible tone. Think of it as a discussion among industry colleagues who are well-versed in the subject matter. Not arrogant or too teach-y, but also not like a couple of teens on instant messaging.
Style: The writing should strike a balance between formality and engagement. Avoid overly casual language, but also steer clear of language that is too stiff or academic. Aim for a tone that would be suitable for a high-level meeting, but not out of place in a one-on-one conversation with a colleague.
Nuance: Maintain a level of professionalism that resonates with an audience familiar with the tech landscape. While the tone should be serious enough for a professional setting, it should also allow for some personality to shine through, so it doesn't read like a research paper.
When I receive an email or see a post thatās, well, just a little too peppy, my sniffer picks up a familiar scent. Hereās one from ChatGPT thatās over-the-top!
Ho-Ho-Hello wonderful supporters and friends! š We at [Nonprofit Organization Name] are absolutely thrilled to invite you to our annual Holiday Gathering! š š Get ready for an evening bursting with festive cheer, heartwarming connections, and a sprinkle of holiday magic! š
šŖ Whatās on the Agenda?
Delicious Holiday Treats & Refreshments š«
Live Music & Carol Singing š¶
Fun Games & Activities for All Ages š²
A Special Gift Exchange š (Bring a gift, get a gift!)
And a Surprise Guest from the North Pole! š
š Dress Code: Feel free to don your most cheerful holiday sweater or outfit! Let's fill the room with color and joy! š
This gathering isn't just a party ā it's a celebration of our community's incredible spirit and generosity. Let's come together to share joy, laughter, and gratitude for the amazing things we've achieved this year. š„³
Calling Card #3: Dudes with Short Beards in Images
Letās turn to image generation now. This is a target-rich environment for AI sniffers. There are still so very many ways that AI-generated images are quirky.
One thing Iāve noticed, both in generating my own images and more profoundly in this issue of Jim Wrubelās hysterical weekly AI Outtakes, is that males in these images almost always sport close-cropped beards.
So now, when I see what appears to be a photo and all the guys have beards, my snifferās telling me that itās more than a coincidence.
Calling Card #4: How Many Fingers?
Again picking on text-to-image, which is absolutely amazing but still has a way to go ā¦ its portrayal of human hands and limbs can be bizarre. (Check out Jimās AI Outtakes issues for many of these!)
So the calling card here can be a hand with six fingers. Sniffer alertādefinitely AI-generated! š²
But what caught my attention about this short-term shortcoming harkens back to my introduction, where I pointed out that with new technology often comes malicious use to exploit it. That same mal-intent can be used to take advantage of tech quirks, like hands with six fingers.
Enter the āsixth finger ring,ā which has been rumored to be of interest to amateur criminals. The story goes that if thereās surveillance video that catches them red-handed (sorry, couldnāt resist!), they can claim in court that the video must be an AI deep-fake, because it shows a hand with six fingers, but clearly theirs has the usual five. (Iām picturing OJ trying on the tiny glove.)
Conclusion
While apps that claim to detect AI come and go, our human sniffers still have the upper hand.
When you use AI to generate content or images, perhaps keep this in mind! Go light on the emojis, tone down the peppiness, and try to create images with some diversity of facial hair but consistency in digit counts!
And above allā¦
Make Good Choices