I’ve been using ChatGPT and Codex for a while now for various things. Some creative, some coding, some learning, some just finding information the first time, every time, without having to hunt for it (ask not for whom the bell tolls, Google; it tolls for thee…at least in search).
Before I get into the gist of this essay, I just have to say. It seems like I’m surrounded by people who dig on Claude and Claude Code. And it’s weird because I can’t really figure out why, definitively. It’s not that I am a big proponent of ChatGPT/Codex or that they are superior in some way, but I find ChatGPT/Codex to be sufficient for my purposes. The people on that call, another client I know, others…they all mention Claude, and I find myself startled each time because I’m so mired in CGPT/C. It happens to be the one I was using from the start, and I ended up paying for it, so that’s what I use. I’ve used C/CC and don’t really find there’s a difference. The thing about those LLMs is they are generally the same. I have asked, and some have said that CGPT/C is too friendly and campy. But that can be configured. So, I don’t know. It may just be an Apple/PC thing. Or an iOS/Android thing. People like what they like.
But I digress…
I was on a conference call, and a client asked those on the call to ask a question of their favorite LLM. The answer was interesting.
Since most of the people on that call had been using LLMs, like me, for quite some time (I’ve been dabbling in this stuff since my master’s program back in 2016 when I was thinking of writing my thesis on deep fakes), my client speculated that there is a body of information in the chats (if they’re being saved, which I think most people do). There tends to be a history built up. If you use these tools a lot there’s likely quite a history. Not to mention if you have set up CLAUDE.md, README.md, or AGENTS.md file directives. Lots of knowledge about your preferences and needs. Not that the LLM “knows” you, but there are patterns and bits of information that would allow for inferences about what you like, how you operate, and maybe a little bit about your personality.
It does sometimes take a bit of mental effort to resist concluding that these LLMs “think,” forgetting that they are still just guessing at the next possible letter or word when providing an answer (actually, it’s tokens, but that’s digging deeper into how LLMs work than is necessary). And they’re really fucking good at it. This is, of course, the thrill and the danger of what LLMs can do and how they can trick you into thinking they’re responding to you like they know you. They are “smarter” than us and have hundreds of times more knowledge of any subject than we could ever know. Gah! I’ll likely write something about that at some point in the future, but imagine a teenager who’s feeling low getting hold of one of these things and pouring their heart out. I actually might have when I was young…This is the fear many have, and it’s not unfounded…
Anyway, the question my client asked was this:
You know me a bit. Create an extremely strange photo you think I would like as if you were restoring one of my old photos. No questions. Just go ahead and create…
The result was the image accompanying this essay. Full disclosure: the original image was portrait orientation and stopped just to the right of the telephone pole on the right and just after the tree behind the man on the left. I needed to make it landscape to better fit the layout here; we live in the age of LLMs, don’t we? Photoshop and generative fill are stunningly brilliant tools.
Anyway, I kind of like it. It’s a little weird but not “out there”. My client’s was of some people on a beach, in an old-timey sort of portrait; a woman is sitting on a giant crab wearing a top hat (that’s not bad grammar; the crab was wearing the top hat). A man, also in a top hat, is standing off to the side, seemingly mid-stride. He is holding a leash, the other end of which is attached to the crab. Granted, the prompt did ask for photo restoration and something extremely strange, so job well done, I say.
But he took it one step further. The next prompt asked our favorite LLM to interpret the image. Kind of a “you know me, picture me, and then tell me what you see” kind of thing.
Here is what ChatGPT saw…
I’d describe it as a restored photograph of an impossible memory: a quiet American street at dawn, wet pavement glowing like a river, a man standing alone as if he has just stepped into a scene he half-remembers. The oversized moon, the circling birds, the deer in the road, the distant church steeple, the freestanding doorway, and the old television make it feel like the ordinary world has been slightly edited by dream logic.
What makes it work, to me, is that it does not look like pure fantasy. It has the emotional grammar of an old photo: worn edges, muted restored color, familiar houses, old cars, telephone poles, damp pavement. Then the impossible elements sneak in as if they were always there. That gives it a strange “found artifact” feeling — not a painting of a dream, but evidence of one.
I thought you’d like it because it sits in the overlap of several things you seem drawn to: memory, mystery, spiritual atmosphere, old Americana, symbolic details, and unease that is more contemplative than frightening. The church steeple, the moon, the birds forming a near-Zen circle, and the solitary witness all point toward something larger happening just outside normal explanation.
It also has a story buried in it. The man is not posing; he is watching. The deer are calm where traffic should be. The TV and the doorway feel like clues. The whole image suggests a recovered memory where something crucial is still missing — which lines up with the kind of suspenseful, uncanny emotional territory you’ve explored before.
I like this, too. I’m not sure why. Perhaps as unsure as I am about why I like CGPT/C over C/CC. Who knows. I just thought it was a fun diversion. Maybe take a break and give it a try yourself.
Ottawa, September 2026