— 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...
No comments:
Post a Comment