"I'm absolutely certain," said the A.I.
Cockiness isn't a good look, especially for an artificial intelligence bot.
One of my Substack friends recently posted this photo.
“A testimony and testament to our humanity despite all odds,” she wrote.
“That image is AI-generated,” I replied.
“How do you know?”
“I’ve never seen the image before, and it would have appeared on my radar decades ago,” I wrote. “The image is too perfect. The physics of one man carrying the other doesn’t make sense. The clothes. The belt. The sign. The lighting. The general vibe. It all says: AI.”
To be fair, my friend has (probably) not experimented with Midjourney, DALL-E 3, Stable Diffusion, Gemini, ChatGPT, and other AI image generators, and I have. Nonetheless, I wondered if I got it wrong. So, I uploaded the “photo” to Google’s Gemini AI, and asked. Here’s our exchange:
I find the conversation with AI most interesting. This is what an entire generation is using to get their information.
Excellent post. The current version of Google Gemini, 2.5 is a step down from the prior version. It lies. And when you catch it in a lie, it doubles down on that lie. It is great for making things up like images and videos. But it worthless when you want facts.
I plugged the image into TinEye reverse image search and it had zero hits for prior images on the internet. Google image search only found matching images from the past few days.
Google Gemini lies.