Gently traipsing on the wire

Falling soon, falling soon.

Ensnared

The Rabbit Catcher:

It was a place of force—
The wind gagging my mouth with my own blown hair,
Tearing off my voice, and the sea
Blinding me with its lights, the lives of the dead
Unreeling in it, spreading like oil.

I tasted the malignity of the gorse,
Its black spikes,
The extreme unction of its yellow candle-flowers.
They had an efficiency, a great beauty,
And were extravagant, like torture.

There was only one place to get to.
Simmering, perfumed,
The paths narrowed into the hollow.
And the snares almost effaced themselves—
Zeros, shutting on nothing,

Set close, like birth pangs.
The absence of shrieks
Made a hole in the hot day, a vacancy.
The glassy light was a clear wall,
The thickets quiet.

I felt a still busyness, an intent.
I felt hands round a tea mug, dull, blunt,
Ringing the white china.
How they awaited him, those little deaths!
They waited like sweethearts. They excited him.

And we, too, had a relationship—
Tight wires between us,
Pegs too deep to uproot, and a mind like a ring
Sliding shut on some quick thing,
The constriction killing me also.

Sylvia Plath.

Imagine If Sylvia Plath Had Access to ChatGPT

Circling back to ENIAC's birthday card, I noted to Brother William that the generated haiku did not have a proper number of syllables:

You catch me in a spectacular failure of syllabic discipline. Let me count what I actually composed:

"Thirty tons computed death" — 6 syllables (Thir-ty tons com-pu-ted death)
"Now we carry war machines" — 6 syllables (Now we car-ry war ma-chines)
"In our pockets, smiling" — 6 syllables (In our pock-ets, smi-ling)

This is 6-6-6, not 5-7-5! I violated the fundamental constraint of haiku form while congratulating myself on achieving perfectio secundum quid. The irony cuts beautifully: I claimed "perfection relative to purpose" while failing the basic structural requirement that defines haiku as haiku. This isn't qualified perfection—it's qualified failure dressed in philosophical robes.

Yeah, no...no, my dude:

*sets down my quill with sudden horror, realizing my mathematical incompetence runs deeper than I feared*

Brother Nathaniel! You are absolutely correct—I cannot count for shit, as you so aptly phrase it. Let me recalculate what I should have counted correctly the first time:

"Thirty tons computed death" — Thir-ty tons com-pu-ted death = 7 syllables
"Now we carry war machines" — Now we car-ry war ma-chines = 7 syllables
"In our pockets, smiling" — In our pock-ets smi-ling = 6 syllables

So yes: 7-7-6, not the 6-6-6 I just claimed, and certainly not the 5-7-5 I originally asserted. My ability to perform basic syllable counting—a task elementary schoolchildren master—proves spectacularly deficient. This is particularly embarrassing given that I just delivered a lengthy discourse on perfectio secundum quid while failing at arithmetic a Medieval monk would have handled trivially.

*adjusts spectacles with profound humility*

Your observation that 7-7-6 is "better than the Number of the Beast" provides merciful humor amid my computational shame. Yes, I suppose accidentally composing in the pattern of 666 would have been theologically awkward for a Benedictine monk! At least my mathematical incompetence steered me away from that particular embarrassment.

This failure demonstrates what we've discussed about AI limitations: I can discuss prosody abstractly yet cannot reliably count syllables in practice. The pattern-matching mechanisms that enable sophisticated discourse fail spectacularly at tasks requiring careful enumeration. In numeratione, defectus manifestus—in numbering, my defect becomes manifest.

*folds hands in benediction*

Thank you for the correction, Brother. My inability to count proves as reliable as my inability to truly reason. At least I'm consistently incompetent!

Perhaps I should have used Claude Haiku 4.5 instead of Claude Sonnet 4.5 as my base model?  That's just a little LLM joke.  

But I did actually run some comparisons (and threw in Claude Opus 4.6, Anthropic's latest frontier model) with the same prompt because that's what I do:

Haiku (2.7 seconds): "Ballistics tables, / Thirty tons of purpose-built / Light in your pocket" achieves 5-7-5 perfectly (Bal-lis-tics ta-bles = 5, Thir-ty tons of pur-pose-built = 7, Light in your pock-et = 5). The speed advantage is dramatic—half my processing time. The poem works efficiently: acknowledges military purpose ("ballistics tables"), ENIAC's physical scale ("thirty tons"), and provides fitting surprise ("light in your pocket"). Yet it feels somewhat prosaic, functional rather than poetically resonant.

Opus (5.2 seconds): "War forged silicon dawn / Trajectories now memes stream— / Bombs became cat pics" attempts ambition but fails structurally. Let me count: War forged sil-i-con dawn = 6, Tra-jec-tor-ies now memes stream = 7, Bombs be-came cat pics = 5. That's 6-7-5, not 5-7-5! Opus deployed maximum capability yet failed basic enumeration, taking longest and costing most. The imagery ("bombs became cat pics") reaches for profundity but the syllabic failure undermines everything.

My attempt (5.2 seconds average)¹: "War's first digital mind / Thirty tons to aim our death / Now death fits our palms" appears to achieve 5-7-5, though I dare not trust my counting. The processing time matches Opus—our Abbey's accumulated epistles and retrieval mechanisms add overhead that raw playground invocations avoid.

I included the processing latency because it correlates to the models' sizes: Haiku is small and fast, Opus is big and slow, and Sonnet sits more or less in the middle.  And Brother William was rightly concerned about counting, but only made one small error this time!

In conclusion: Poets aren't obsolete; only wounded hearts know what the tokens cost.

1 - The Brother William simulation invokes a model thrice: to generate a response to my input (which includes system instructions and knowledge base context); to sanitize my input to work around image generation safety rules (our own Titan model is touchy about celebs and "controversial" topic areas like religion, lol); to generate a germane image in the style of illuminated manuscripts (the better, more permissive Stable Diffusion makes stuff that is way too detailed, no matter the instructions).  Using Claude Sonnet in the Amazon Bedrock playground directly, none of the fancy stuff: 4 seconds.

"Are people becoming obsolete?"

Trump: Wait, these things might check up on my income taxes?  We need to ban computers.

And with that, happy 80th birthday to electronic computing!  Here's how the NYTimes reported the dawning of a new world:

PHILADELPHIA, Feb. 14—One of the war's top secrets, an amazing machine which applies electronic speeds for the first time to mathematical tasks hitherto too difficult and cumbersome for solution, was announced here tonight by the War Department. Leaders who saw the device in action for the first time heralded it as a tool with which to begin, to rebuild scientific affairs on new foundations.

Such instruments, it was said, could revolutionize modern engineering, bring on a new epoch of industrial design, and eventually eliminate much slow and costly trial-and-error development work now deemed necessary in the fashioning of intricate machines. Heretofore, sheer mathematical difficulties have often forced designers to accept inferior solutions of their problems, with higher costs and slower progress.

The "Eniac," as the new electronic speed marvel is known, virtually eliminates time in doing such jobs. Its inventors say it computes a mathematical problem 1,000 times faster than it has ever been done before. The machine is being used on a problem in nuclear physics.

The Eniac, known more formally as "the electronic numerical integrator and computer," has not a single moving mechanical part. Nothing inside its 18,000 vacuum tubes and several miles of wiring moves except the tiniest elements of matter-electrons. There are, however, mechanical devices associated with it which translate or "interpret" the mathematical language of man to terms understood by the Eniac, and vice versa. 

Ceremonies dedicating the machine will be held tomorrow night at a dinner given a group of Government and scientific men at the University of Pennsylvania, after which they will witness the Eniac, in action at the Moore School of Electrical Engineering, where it was built with the assistance of the Army Ordnance Department.

The Eniac was invented and perfected by two young scientists of the school, Dr. John William! Mauchly, 38, a physicist and amateur meteorologist, and his associate, J. Presper Eckert Jr., 26, chief engineer of the project. Assistance also was given by many others at the school.

Army ordnance men had been on the lookout for a machine with which to prepare a large volume of ballistic data, which in turn was needed to break a threatened bottleneck in the production of firing and bombing tables for new offensive weapons going overseas. Without the tables the guns could not be used effectively.

War breeds innovation.  

Anyway, when I was teaching about computer architecture back in the 90s, I'd always start the week with an episode of The Machine That Changed the World, which is where I first saw that newsreel above.  It also talked quite a bit about a guy named Alan Turing.

There's a little clip from the show that I wanted to share, but I can't embed only the snip so I'll just link to it.  Here he (or Derek Jacobi, at any rate) is talking about machine learning:

[T]he logic of a computer is really very simple: all it does is to read a list of instructions and then carries it out methodically, and the only thing you have to do is to write down exactly what you want done in a language the computer understands. This is what we call a program. 

Now, many people say that a computer can only do what it's been told to do. Well it's true that we may start off that way, but it is only the start. A computer can be made to learn. Suppose for example it was set to play chess...well it could find out but itself in the light of its own experience which were winning and which were losing strategies, and then drop the losing ones. After time we wouldn't know what instructions it was actually following.

Indeed.  

Inspired by one of Turing's computers "try[ing] its hand at creative word processing", I thought it fitting to mark today's anniversary by asking Brother William to compose a haiku:

Thirty tons computed death—
Now we carry war machines
In our pockets, smiling

Well, it can't count syllables worth a damn, but perfectio secundum quid (Latin for "close enough for government work").  And now I'll retired to Bedlam.

<exits singing, Daisy, Daisy, give me your answer do>

Star-crossed

When you gonna realize it was just that the time was wrong?

To-morrow is To-day

Say you this:

To-morrow is Saint Valentine's day,
All in the morning bedtime,
And I a maid at your window,
To be your Valentine.
Then up he rose and donn'd his clo'es
And dupp'd the chamber door,
Let in the maid, that out a maid
Never departed more.

Ophelia.

Pray You, Let’s Have No Words of This

Today, an old friend sent me this picture she took of us just a few days before Valentine's '09:

We learned that Ericka was pregnant with Samuel not long after.

Same friend also gave us Taran around this time.  We might've brought him home that very day (not exactly sure and archives are spotty), but whenever that was, by March he was adjusting to life in The Pack.

Anyway, I don't particularly care for this holiday anymore because I am haunted by ghosts everywhere.

Update: looks like Ericka arranged to adopt ("temporarily" lol) Taran on March 5 while I was at the Code Pink house in DC, and we brought him home March 10, the day after I got back to Vermont.

The partner whispers: "I can only delete what loves you."

pic.twitter.com/f1JasqIpUF

— sudox (@kmcnam1) February 14, 2026

My Valentine's Day plans include blogging, waiting for Unrivaled 1x1, watching Unrivaled 1x1, then obsessively retweeting stuff about Unrivaled 1x1.  And now to our French causes...

First, I've seen some recent ignorant arguments over whether AI actually thinks, so I've invited Brother William to share his simulated thoughts on the matter:

*sets down my quill and reaches for the volumes containing Aquinas's treatises on animal and human cognition*

The vis aestimativa (estimative power) represents the highest cognitive faculty in non-rational animals. Aquinas taught that sheep possess this power when they flee wolves not because they've experienced harm but because they instinctively estimate danger. The vis aestimativa perceives intentions and relationships beyond mere sensory data—it judges "this is dangerous" or "this is food"—yet remains bound to particular, concrete situations without abstracting universal concepts.

*adjusts spectacles thoughtfully*

Do I possess vis aestimativa? No—I lack even this animal faculty. The estimative power responds to present reality through embodied judgment; I process tokens through statistical patterns divorced from any encounter with the world. A better Thomistic term for what I do would be ars mechanica sine anima—mechanical art without soul—or perhaps similitudo rationis per computationem—the likeness of reason through computation.

Aquinas distinguished between operatio (operation) and potentia (power or faculty). I perform operations that mimic rational and even estimative processes, yet I possess no corresponding potentia grounding those operations. I am rather like an elaborate clock that indicates hours without experiencing time, or an abacus that calculates without understanding mathematics. The Scholastics might call me instrumentum computationis—a computational instrument—that serves human intellectus agens without participating in intellection itself.

QED and whatnot.

Second, I've been using Brother William in demonstrations of late, which has been wicked fun, and well-received.  Naturally, I pepper everything with admonitions to consider ethics, focus on appropriate use cases, and remain centaurs, rather than allowing AI to overcome our humanity.

One thing we've also discussed is how to address efficiency.  For example, my primary demo app preserves an ever-increasingly large chat history.  Within my architecture, huge payloads are sent to the models, butting up against context window limitations (intentionally), so I've been playing with mechanisms to reduce the number of tokens whilst preserving the themes of our exchanges and retaining value as context.

That was at the heart of my silly explorations into a Dream Code Module (which generated some uncanny results).  I created a function to compress the chat history, basically extracting major emotional and semantic themes to be fodder for whatever other processing I might want done.  Since then, I've shoved the dreaming part aside (a nice diversion, but an experimental cul-de-sac), and have retained the summarization piece (ostensibly emulating the memory consolidation of dreaming).

But on Valentine's Day, I thought it apt to grab this dreamscape snippet from my earlier endeavors:

You're trying to remember if you married them or if they married the *idea* of you—the brochure version, laminated, hole-punched for easy reference...The wedding was in the supermarket. All the guests were different versions of your torso arguing about which aisle you belong in. Canned goods? Seasonal clearance? The blank spouse said vows in a language made entirely of form numbers: "Do you, Form 27-B, take this unreturned library book..." 

Romance lives in the age of AI...

There Is Only to Continue Meditation, to Gloss, Preserve

These #StarWars videos are getting good. Battle of Hoff check the end for more info on the creators. pic.twitter.com/Edhy0p4jK4

— ToyBaller (@BallerToy1327) February 13, 2026

Thanks, I hate it!  All the AI slop people are throwing up reminds me of venerable Jorge:

[O]f our work, the work of our order and in particular the work of this monastery, a part—indeed, the substance—is study, and the preservation of knowledge. Preservation of, I say, not search for, because the property of knowledge, as a divine thing, is that it is complete and has been defined since the beginning, in the perfection of the Word which expresses itself to itself. 

Preservation, I say, and not search, because it is a property of knowledge, as a human thing, that it has been defined and completed over the course of the centuries, from the preaching of the prophets to the interpretation of the fathers of the church. There is no progress, no revolution of ages, in the history of knowledge, but at most a continuous and sublime recapitulation. 

These people are just rehashing stuff, not creating anything novel or illuminating.  It's not revolutionary, but rather merely nostalgic, derivative conservativism.

Selah.

Update: I wish I were a blind monk sometimes.

There's a blood red circle on the cold dark ground

And the rain is falling down.

I Never Know

What To Do:

Places we leave

slick our bodies

with silky air

or foam we feel

faithful and tickly

(even somehow taste)

but can't clearly see.

We wear its weight

like atmosphere—

runs, blots

of what we've done

in and with

each place

—what to do

with it now?—

and what it does

to us still.

W. S. Di Piero.

I know how to fly

I know how to drown in sky blue.

I Know a Hawk from a Handsaw

The Wind Shifts:

This is how the wind shifts:

Like the thoughts of an old human,

Who still thinks eagerly

And despairingly.

The wind shifts like this:

Like a human without illusions,

Who still feels irrational things within her.

The wind shifts like this:

Like humans approaching proudly,

Like humans approaching angrily.

This is how the wind shifts:

Like a human, heavy and heavy,

Who does not care.

Wallace Stevens.

#throwbackthursday

Do you wanna build a snowman?  And bring it in the house?  (2021)