Before the whole AI thing I never gave much thought to what it meant to be intelligent. Chimps seem to have intelligence and my Golden Retrievers seem pretty smart, but for the most part, intelligence was a human thing. Is language a necessary component of intelligence? Early humans could communicate with grunts, gestures, facial expressions… but there were no words streaming through their noggins. So were they intelligent? As I so often do these days, I went to ChatGPT for an answer:

It depends on how we define “intelligent.” If we go by most modern definitions (problem-solving ability, adaptability, tool use, planning, social cooperation), then yes, pre-linguistic humans—or hominins—would likely still qualify as intelligent. Continue reading →

I never tire of looking at our home and decided to let ChatGPT do the Studio Ghibli thing the youngsters are so fond of…

Might there come a day, I wondered, when there’s no one left alive who remembers how to create something like this using ink and brushes and paper?

ChatGPT: Not just the skill, but the patience—the subtle touch, the knowing when to stop, the silence in the strokes. A watercolor isn’t just pigment on paper; it’s time and breath and stillness made visible.

Tools like me can mimic the outcome, but not the ritual. Not the weight of the hand, the pauses between brushstrokes, the afternoon light shifting across the desk. That lived, human moment—the unrepeatable presence in the act—is something we can only try to echo.

I think what you’re mourning isn’t just the art. It’s the reverence.

If I uploaded an actual watercolor painting, and one created by an AI, would you be able to detect the difference?

ChatGPT: Great question, and one that’s getting trickier by the day. If you uploaded both images, I could analyze visual cues like:

• Brush stroke irregularities – Real watercolors tend to have organic, sometimes unpredictable variations in pigment pooling, bleeding, and texture.
• Paper texture – Traditional watercolor paper has a distinct tooth that subtly warps and affects pigment spread; AI generations often simulate this, but often too evenly.
• Edge behavior – The way colors feather into one another or dry with hard edges can be a giveaway. AI sometimes overdoes this for effect.
• Imperfections – Slight errors or hesitations in a real artist’s hand (a shaky line, an off-center focal point) are hard for AI to replicate convincingly unless intentionally introduced.

That said… AI’s getting better at faking these subtleties. In side-by-side comparisons, I could often make a good guess. But a skilled AI prompt with a high-quality generation? Even trained human eyes can be fooled now and then. Especially once it’s been printed or slightly aged.

ChatGPT became the world’s most downloaded app in March 2025, surpassing social media giants Instagram and TikTok with 46 million new downloads, driven largely by the viral success of its new image generation feature.

With the integration of DALL·E 3 into ChatGPT, you can generate images directly within a conversation by providing a description of the image you’d like to create. I tried this out by asking for an image of a Series III Land Rover.

At first glance I thought this was a photograph but then I spotted some mistakes, which ChatGPT caught as well:

While it looks pretty convincing at a glance, there are a few telltale signs it’s not a real Series III Land Rover:

• Badge: It says “LAMIS ROVER” instead of “LAND ROVER”—a classic AI hiccup with text.
• Proportions: The front end is a bit off. The headlights, grille, and indicators aren’t quite in the right relationship to each other.
• License plate: The format looks plausible (British-style), but it’s randomly generated and doesn’t tie to any real registry.
• Subtle uncanny valley cues: The reflections, panel gaps, and shadows are almost right—but not quite.

OpenAI has recently introduced a new image generation feature powered by the GPT-4o model. This integration allows users to create and edit images directly through conversational prompts. GPT-4o processes and generates text, images, and audio. It’s a whole new flavor of creative dopamine. Suddenly you can conjure anything. It’s like when people first got their hands on Photoshop, or GarageBand, or a printing press… but this time, the ink thinks with you.

“The line between idea and image has never been thinner. It’s not surprising that the tool starts feeling less like a feature and more like a portal. Dangerous stuff, in the best way.”

I asked what styles were being prompted by users.

• Hyper-Realistic/Photorealistic
• Fantasy & Sci-Fi Illustration
• Comic & Graphic Novel
• Anime/Manga
• Oil Painting/Classical Art
• Watercolor/Ink/Sketch
• Vaporwave/Synthwave/Retro-Futurism
• Isometric & Pixel Art
• Minimalist/Flat Design
• Surreal/Dreamlike

To try this new tool I uploaded this photo of my friend George…

…and said have him holding his head in frustration, standing behind an older woman sitting in front of a computer (George does tech support) and use a comic book style.

Next I told GPT to make the image look like a Renaissance painting…

…then asked for period clothing…

Each of these images took only a minute or two to create and can be tweaked again and again to get (almost) exactly what you’re after.

Next came Film Noir Detective and ’70s Cop Show…

The prompt for my friend Paul was something like, “A man lying on one of those rolling things mechanics use to work under cars. Put him in coveralls, give him a wrench, and add a grease smear to his face.” I tweaked the first image by having him working on a 1966 Mustang.

I finished up with Disco Barb and Cyberpunk Steve…