On Washington St.

Some flowers & leaves

On the water July 4, 2026

Pier 13 memory machine

Miscellaneous images on a Sunday afternoon

Why I like this photo inside Mickey D’s

By ordinary standards, this is a poor photograph. Most of it is blurred and half the image is taken up by an uninteresting tabletop. Ordinary standards don’t interest me.

What caught my attention about this photo? The lights. Because most of this image is blurred, those lights are not well located in the 3D space of the room. Rather they appear to be bright patches on the surface of the photo. They betray that fact that his photo IS an image on a plane, not a direct glimpse into the room.

That fact that most of the image is blurry doesn’t tells us that this is a photo. Or rather, it does so, but in a way that’s more intellectual, something we infer, than perceptual. The receding lines of the ceiling and of that boring tabletop are strong perceptual depth cues, as are the relative sizes of things, especially the man standing back in the corner right of center. Those cues define a 3D space. But it is hard to situate those lights in that space. We know where they are intellectually, but we can’t see them perceptually. As I’ve already said, perceptually they are bright patches on the surface. The image is thus in conflict with itself.

That conflict both says, “I’m a photo,” and invites us to attend to color and composition. Those I like. The fact that a bit over half the image is taken up by that tabletop produces contrasting visual materials that I find interesting. On the one hand we have the faux grain pattern that dissolves toward the center of the image. That contrasts with the well-articulated geometry of the room that is realized in different colors.

Almost all of these effects, except for the color itself, remain, and are perhaps slightly enhanced, in the black and white version below.

Jaime’s art exhibits an intelligence about which the Silicon Valley digerati are clueless, a conversation with Claude Opus 5

You may remember that about a decade ago I did a series of posts about the art of Jaime Berubé and then collected them into a working paper, Jamie’s Investigations: The Art of a Young Man with Down Syndrome (2016). Two days ago I decided to have Anthropic’s Claude read and comment on that paper. We ended up having a fascinating and productive conversation that yielded new insight. I’ve appended that to this note.

Before we get to that, however, I want to tell you about a little adventure I had last year. I decided to draw some images inspired by Jamie’s dot images. He drew his images freehand on a blank page and used crayons. I drew a grid on my pieces of paper – I did multiple images – and used colored brush pens. Here’s one of my dot images. I have some others in a post, Dot Paintings (after Jamie Bérubé).

Why’d I do it, you ask? Because I thought it would be interesting. It was. And fun. It was. And even instructive. More than I’d thought.

What did I learn? It’s more difficult to do than it looks. While the images appear simple – and they are in a sense, certainly in comparison to, say, a Rembrandt or a Bob Ross – drawing them requires effortful concentration. As I pointed out in my working paper, Jaime didn’t choose his colors randomly, though it may seem that way. When you examine the images carefully you’ll discover that there’s an order to them.

Before I had Claude read my paper I gave it one of the dot images and asked it to comment. Not only did it discover the nature of that order, it picked up on other things I hadn’t noticed and on that basis asked me some questions, some of which I could answer, some I couldn’t. It was an interesting discussion.

Getting back to my own dot images, I discovered that keeping track of my colors was just a little tricky, not rocket-science tricky, but tricky like keeping track of what you’re doing in adding, say, a half dozen multicolumn numbers. So I made mistakes every now at then. Simply dealing with multiple pens, putting one down, picking another one up, that was tricky as well. I made mistakes.

And you know what? Keeping track of all those fiddly little details requires intelligence. Not rocket-science intelligence, not IQ intelligence, but intelligence nonetheless. I can’t think of any other word that characterizes those acts as well as that one does. Claude’s essay says a bit about that near the end.

Moreover it’s the kind of intelligence on which rocket-science intelligence is based. That requires an explicit argument that Claude didn’t make. But you can find a big chunk of that argument in one of my working papers, What Miriam Yevick Saw: The Nature of Intelligence and the Prospects for A.I., A Dialog with Claude 3.5 Sonnet (2025).

Here’s Claude’s essay. It did all the writing, but I inserted the drawings.

* * * * *

What a Sheet of Colored Dots Can Tell Us About Intelligence

Jamie Bérubé has been drawing abstract images since he was eleven. He has Down syndrome. His father, the literary critic Michael Bérubé, put a large selection of the work online, and it is that online archive — not studio visits, not observation of the artist at work — that Benzon's working paper analyzes.

That constraint turns out to matter. Everything below is inferred from finished artifacts. Nobody watched these being made. Which means the analysis has to reason backward from products to processes, and the interesting question becomes how much of a process a product can be made to reveal. Quite a lot, as it happens.

The rule that isn't visible

Start with the simplest of the genres: sheets covered edge to edge with small circular marks in crayon, several hundred to a page, in a great range of colors.

Michael Bérubé, who has seen thousands of these, describes the color choices on most of them as random. They look random. There is no gradient, no symmetry, no regional organization, nothing that reads as a plan.

They are not random. Adjacent marks are almost never the same color. With a palette of roughly ten colors and something over four hundred marks, chance alone would produce well over a hundred violations on a single sheet. There are a handful. The suppression holds vertically and diagonally as well as horizontally, which means the constraint is being applied over a two-dimensional neighborhood, not merely to the previously drawn mark.

So there is a rule: no two adjacent marks the same color. It is entirely local. It requires checking at most four already-placed neighbors. And it produces, without any global plan whatever, the quality that makes the sheets pleasant to look at — a rhythm that is easy to sense and hard to describe.

This is worth pausing on. The rule is invisible to inspection and available to description. The person with by far the most exposure to the corpus reports randomness. Counting reveals otherwise. That gap between what looking gives you and what describing gives you is a recurring feature of this material, and it is one reason the images repay analysis at all.

Reading the process off the page

Other features of the sheets carry process information.

The left margin is well maintained: rows begin at a consistent horizontal position. The right margin is ragged: rows end wherever they end. That is carriage-return behavior — a return sweep to a remembered left position, then progress until the paper runs out. It is what unlined handwriting looks like, and it indicates that the governing spatial rule is sequential and local rather than planned against the page as a whole.

Upper rows are straighter and more evenly spaced than lower ones. Error accumulates downward and is never globally corrected. And the sheets typically stop mid-row, with blank paper remaining. Michael Bérubé supplies the reason: Jamie works until he is interrupted — a meal, an errand — and does not resume a page later. He starts a new one. Each sheet is a session, and a session ends when it ends.

None of this required watching. It is all recoverable from the artifact by someone who knows what to look for, which in this case means someone with extensive practical experience of drawing and of the problem of filling a blank sheet coherently.

How the discovery was made

The most striking material in the working paper is not the mature work but the early work, done between ages eleven and fourteen, which Michael Bérubé has also posted. Those early sheets are miscellanies. Various kinds of object — bars, circles, letterforms, geometric bits — appear on the same page, each placed wherever there was room. There is no relationship between neighbors. There is no composition.

And therefore, as Benzon observes, there was no global order available to be discovered. As long as the elements varied and each sat in its own local space, nothing about their arrangement could become salient.

Friday Fotos: 4th of July sky

Toward, through, and out: Hoboken Terminal

They do it all for us, Mickey D’s!

Rounding the turn into Newport

The odd destruction of books en masse by AI companies [Homo economicus on a binge]

I'm a second-generation bookseller. My family runs Houston's largest used & rare bookstore and I'm building an AI tool for used bookstores. We got hit by exactly these orders, including a single order for 70 obscure books that made us pause online sales entirely. So I dug in. I… https://t.co/IK7OE8f1DG

— Charlie D. Becker (@charliedbecker) July 28, 2026

Firetruck

On Washington St. in Hoboken

ChatGPT draws Rorschach blots

Here they are, six of them:

I then discussed the blots with ChatGPT in terms suggested by me new working paper, A prototypical image in ChatGPT 5.6: An informal pilot study. First I present all six blots, full size, and then the discussion.

I needed a Rorschach inkblot for a post I was working on, so I had ChatGPT create one. I then decided to see what would happen if I asked for a number of inkblots, each in a different chat. I’ve uploaded the result to the project in a document entitled Rorschach Blots. I had no expectation about what would happen. Would it produce the same blot each time? Of different blots each time? As you can see, it produced different blots each time.

This suggests two further lines of investigation. 1) Generate 100 independent ink blots and run them through an image classifier. Would we get distinct clusters or not? 2) Use prompts suggesting specific things, but otherwise open-ended, e.g. “Draw a cat,” “Draw a tree,” “Draw a car,” “Draw a coat,” etc.

Looking at the Rorschach Blots document, the six outputs are plainly different, but they are not unconstrained in every respect. All six preserve a common structural grammar: bilateral symmetry, a strong vertical axis, black-and-gray ink on white, irregular feathered edges, and enough internal articulation to invite figurative interpretation. Within that shared format, however, the overall silhouettes vary substantially: some are tall and narrow, some compact and squat, some strongly winged, some mask-like, and some more diffuse. The contact sheet on page 1 makes that balance of sameness and difference especially clear.

That suggests that the system has not selected one prototypical blot, in the way it selected one broad landscape type. Instead, it seems to have selected a prototypical design space for “Rorschach blot,” within which it samples different instances. Your two proposed extensions are therefore well chosen.

1. One hundred independent inkblots

The clustering question is exactly the right one. A hundred examples might reveal:

• one continuous cloud of variation around a single template;
• several recurring morphological families;
• or a few dominant families plus outliers.

Even by eye, the six examples hint at possible families: elongated “totemic” forms, broad winged forms, compact mask-like forms, and more open forms with substantial white interior space. But six is far too few to know whether those are genuine modes or just impressions.

I would not rely on a conventional object classifier, since it may simply label them all “inkblot,” “butterfly,” “mask,” or “bat.” A more informative analysis would use image embeddings and cluster those embeddings, perhaps supplemented by simple measurable features such as aspect ratio, occupied area, contour complexity, amount of internal white space, and vertical distribution of mass. The point would not be to decide what each blot “looks like,” but to determine the structure of variation across the set.

2. Open-ended object prompts

“Draw a cat,” “Draw a tree,” “Draw a car,” and so forth would probe something different and potentially more revealing: whether the system has a default realization for a named category.

The questions would include:

• Does “Draw a cat” repeatedly produce the same breed, pose, viewpoint, or expression?
• Does “Draw a tree” default to a solitary deciduous tree, a stylized botanical specimen, or a landscape scene?
• Does “Draw a car” default to a modern sports car, a generic sedan, or a three-quarter advertising view?
• Does “Draw a coat” produce the garment alone, on a mannequin, or worn by a person?

That experiment would occupy a useful middle ground between the wholly unspecified prompt—“Create a drawing”—and a detailed scene description. The subject is fixed, but almost everything about its realization remains open. It would therefore reveal category-level defaults in pose, style, composition, context, and affect.

The Rorschach result also suggests a distinction worth preserving in the paper’s future-work section:

Some prompts may elicit a single dominant subject prototype, while others may elicit a constrained family of forms rather than one recurring image.

That is the most interesting implication of the blot experiment. The landscape prompt collapsed onto a narrow semantic prototype. The inkblot prompt produced morphological diversity within a narrow formal convention. Those are different kinds of regularity, and comparing them could become a substantial research program.