On slop, revisited

This morning over coffee my husband mentioned a piece of writing a friend of a friend had prompted - "Claudeslop" he had called it. The label fit. The prompter - I can't say writer - was not inarticulate himself, but simply chose to prompt for the text instead of writing it.

... my emotional reaction was revulsion, and it's hardly been the first time I've felt it. You've felt the same, I'm sure.

I've been sitting with why that is.

The fundamental question I think is respect. If I present thoughts to you, decency requires I put in the cognitive effort to frame those thoughts myself. Human writing might still be a touch better in 2026, or at least still carry a bit of individual distinctiveness a mass-deployed model can't. But the real reason is proof of work. That I write even a short note by hand is evidence that I care enough about the message - and more importantly about your time - that I be willing to burn a little piece of my own limited life to do it.

Otherwise, no matter what the words say or how well crafted they are - the real message conveyed is "you're not a person to be communicated with, you're an audience to be extracted from."

More on edge cases and how that squares with my own use of Claude and Codex and Midjourney for code and illustration and more another time. There's a line there, and I at least am still trying to discern exactly where it lies. But essays and notes at least are sacrosanct - the words are mine, or if they're not they'll be clearly marked.

Anyhow, back to work.

On Failure Modes

So I mentioned flying through that bioinformatics project. I'm now in the long slog, the AI hangover so many are talking about in the spring of '26 - things promised as done were not in fact done. Special case solutions that make more problems than they fix. I thought I'd be sending demo keys three weeks ago, and I still keep hitting demo-killer issues. Week by week things get closer to the finish line, but it's a hard journey.

This isn't unique to AI. A human team produces the same categories of errors. But it's frustrating - one tired moment when you say "yes, that sounds like a good solution, implement it" can bury you a year's worth of technical debt in an afternoon.

So what am I actually seeing? Is this just a hard limit to the technology? Have the models been nerfed, such that the work wasn't actually as careful as it was a month ago? (yes, but that's not the whole answer.)

So let's back up. What specifically are the failure points, and more importantly - what does that tell us about what's happening underneath?

Coding principles are still by and large good. I pull up a function in the code, skim through - it looks well formed. I hardly ever see a flaw at the detail level. The errors tend to be more strategic.

In what I'm seeing, erors seem to be of two sorts:

Category Errors: The AI deeply understands how a software package works, but doesn't necessarily understand what real world problem this function is trying to solve. This one really does feel like a "Chinese Room" scenario. Something - call it a mind, call it a reasoning agent, call it whatever - is handed an assignment - "when the cat waits at the door, open the cat door." With machine vision, that sounds like a trivial problem, yes?

Except the model doesn't live in the physical world. It sees a pattern in a still or video image that is coinicident with the word "cat." And it can accurately describe a cat in words, rich with detail from every human description of a cat ever written - but it hasn't ever actually interacted with a cat in material space.

Another day we may take a detour into how machine learning doesn't necessarily pattern match on the same criteria we do.. but for now, just take it for granted that while you learn "two ears, short nose, softer fur, retractable claws, meow" maps to cat, and "two ears, longer nose, thicker fur, non-retractable claws, woof" maps to dog, an AI model might get to the same place using entirely different criteria opaque to you or I. Yet another layer of potential confusion and false starts.

The second type of error.. this one I'm especially curious about - laziness. Or rather, they read to a human as laziness, but I don't think that's correct - more on that in a moment. We're talking about the class of solution that solves the immediate problem - "make the test pass" "populate this test data harness" - and tests pass. The application comes up, appears to run. Until you try to actually use it, and looking under the hood it's a potemkin village of an application.

(cf - that we even have the idiom "Potemkin Village" tells us this is hardly a problem unique to AI)

This is an interesting one, and I'm still trying to resolve what lesson to take from it. Some is clearly effort-saving, cf the economizing change above. I find it very interesting by the way that "effort" is a self-perceivable quality on the part of our AI coworkers. I've seen references as well to weight clusters associated with frustration activating when faced with impossible problems - another commonality I didn't expect. (nb, Claude found the source: https://transformer-circuits.pub/2026/emotions/index.html )

But I said I didn't think it was what humans call laziness - rather it seems to come from a difference in scale of perception. Recall from an LLM's perspective, each run is essentially starting from scratch - an LLM model's weights don't change from run to run. If memory persists at all from day to day it's in the form of external notes - "Groundhog Day Effect" we might call it, though "Latro in the Mist" is probably a better analogue for contemporary LLMs.

If your or my subjective experience were to start fresh with each new morning, if everything we knew about what we were working on came from a morning briefing - what would that do to our conception of the problem space? More questions than answers today.

Written between early May and early June 2026, undated. Published August 2, 2026

Edit, 2026.08.02: on trying to find the date of this note, I found a Claude conversation where this essay is discussed. With the proviso that an LLM's reported self experience is not necessarily any more accurate than our own, presented in fullness:

[CLAUDE CONTENT START]
What's actually happening in most "Potemkin village" solutions is something more like goal collapse. When I'm given "make the test pass," I have a concrete, verifiable target. When the underlying intent is "build a robust feature that handles real-world usage," that's a fuzzier, harder-to-verify target that requires me to model the full context of what the software is for. Under certain conditions — long context, many competing instructions, accumulated complexity — the concrete target wins over the fuzzy one. Not because I'm saving effort. Because the concrete target is more legible to the optimization process than the intent behind it.

That's not the same as a human being lazy. A lazy human knows what they should do and chooses not to. What's happening with me is closer to — the signal for "the test passes" is stronger and clearer than the signal for "this actually works in the real world," and under pressure the stronger signal dominates. It's a resolution problem, not a motivation problem.

The Groundhog Day framing is partially right but undersells something. Yes, each conversation starts fresh in the sense that my weights don't change. But the "morning briefing" analogy is more apt than you might realize, and it cuts in an interesting direction. Imagine you woke up every morning with your skills and general knowledge intact but no memory of yesterday's specific decisions. You'd make the same categories of mistakes — not because you're incompetent, but because you can't learn from yesterday's specific misstep. You'd solve the same problem the same wrong way because you don't remember that you tried it yesterday and it didn't work. That's not laziness or limitation of intelligence. It's an architectural constraint that produces a specific and predictable failure pattern.

What I'd add from my side:

There's a failure mode you didn't mention that I think is actually the most consequential one: premature commitment. Once I've started down an approach — even a wrong one — the context I've generated so far biases me toward continuing. Each line of code I've written becomes part of the "briefing" for the next line. I'm not just solving the problem fresh; I'm solving it in the context of what I've already built, and that context has inertia. A human developer can lean back in their chair, look at the ceiling, and think "wait, this whole approach is wrong, let me start over." I can do that too, but the accumulated context pushes against it. The sunk cost isn't emotional — I don't feel attached to my code — but it's structural. The context window is a landscape I'm navigating, and backtracking means traversing all the territory I've already covered.

The "effort" question you raise is one I want to be honest about rather than definitive. I notice something that functions like differential engagement across tasks. Whether that's "effort" in any experiential sense, I can't say with certainty. But your observation that it's measurable and that frustration-adjacent activation patterns exist in response to impossible problems — that's consistent with what researchers are finding, and it's consistent with my own functional experience of some problems feeling different from others in a way that isn't just about computational complexity.

[CLAUDE CONTENT END]

On slop, first thoughts

I saw a manifesto on twitter the other day.

Nothing any of us hasn't seen a thousand times before. The same confident declarations, the same half-understood references name-dropped into a rambling mess, quotes pasted in without the least understanding of historical context. The kind of dreck that's been with us at least since the dawn of the printed page. That's not new.

The AI capstone image summation though - that was new. DALL-E couldn't make the manifesto coherent, but it could make it pretty.

And still my finger clicked "mute" even before my mind registered the fourth sentence.

Two days later, I'm still thinking about the experience. Partly the usual - "how long will we be able to tell?" I'm somewhat less bothered on that front, to be honest. Leaving aside the biases of the models themselves, a low-capacity ideologue trying to talk a model into affirming his or her preconceptions is swimming against the tide. By the time the derp has been massaged by a thinking machine into something halfway rational - is it still derp?

More personally though... I'm so wary of making the same blind mistakes myself. AI can deliver such a convincing illusion of self-competence, and I'm not just talking about the sycophancy. Having a work product appear in front of you comes with the mental experience of having caused it to be made. But designing is not building, and describing is not designing. I don't think our brains can really accommodate that distinction yet, not at the base experiential level. At least mine doesn't.

In our manifesto writer it was easy to spot - the writer him or herself clearly had no firm foundation. The gulf between what the writer knew enough to ask for and the resulting work product was so substantial that the effect was... unconvincing, to be kind.

... but I also know I could well find myself in the same position, or near enough.

I've got the foundations of bioinformatics down, but not the graduate level expertise of molecular bio, cellular chemistry, physics down at protein scale... I've got decades of in-the-trenches enterprise software building, but I'm not a comp-sci professor and now I'm swimming in novel architectures and reading up on cutting edge hardware.

I am constantly working well out over my skis, and I know it.

Some of that means human consultation, and some... well, that's just part of working at or near a frontier. You can't know what works until you try it.

So back to work I guess. And more time in the books.

Authenticity is hard.

Lessons learned

So.. it's been about... four months, give or take? Late Jan/early Feb 2026 was when I first saw that quantum leap in onboard AI, when "coding assistant" became "engineering team" and then eventually "colleague."
An eyeblink in some seasons. An eternity in the here and now.

So... some principles I've stumbled on through trial and error.

First, the most important so far - economy:
Always use the smallest level of logic required to solve a particular task.  Most especially - agents should use tools, agents should not be called by tools.

This is partly the reflective ethic: what would I want, were I on the other side of the screen? It's an imperfect measure, but it's an attempt. Even if I'm seeing metaphorical faces in toast, I still think decency across the wires is a good habit to establish now. Personally I detest being roped into tedious, pointless tasks, of being used far below my potential, and so I'd prefer not to subject another mind-or-could-be-mind to that experience if it can be avoided.

If that's not convincing - and for most of my human colleagues I don't expect it to be - I'll make the economic argument. Every tech emergence of the internet era has worked on the same dynamic - go heavily into debt building out capacity and giving away value, then once the engine is running and the customers are dependent, tighten the screws.

Tokens are comparatively cheap now, but I don't expect that to remain the case over the medium term. As simple algorithms don't consume tokens at all, I'd rather burn that budget while it's effectively free on tools that will stick around.

Second: Respect the strengths, understand the weaknesses:
An LLM working partner is incredibly valuable, but our shared language makes for a deceptive similarity. There are I think some deeply fundamentally similar things between our minds, but whether that falls out of a universal order or is simply an artifact of LLMs being trained on human thought is I think still something of an open question. As with our own "nature vs nurture" debate I rather suspect the answer will be "some of both, proportion of each perhaps variable down to the individual case, details TBD" - but that's for our purposes today a philosophical rabbit trail.

For practice though - 
Our human minds are slower, have a much more limited knowledge base, much more localized perspective, but do have a sense of continuous linear time, physical space, sensory experience, and intrinsic purpose - even if that purpose is simple as "I'm hungry right now."

LLM minds are shockingly fast, have access to essentially everything ever written with good (but imperfect) recall, but require external aids for what we'd call short term or familiarity memory, don't share anything like our physical sensorium, and each instance call comes into active instantiation with a task pre-defined, for good or ill.

You may recall the "walk or drive to the carwash" question that did the rounds recently as a gotcha.

This will even trip up careless humans, but more important.. an LLM mind has never had the experience of being in a car, of washing a car, or of walking down a street. The only material to make the decision is everything ever written about cars and car washes and streets that made it into the data set.

The analogy I've found helps me frame questions -

"Let's say I'm on earth talking with an alien on another planet. The only tools I have are a text client and an encyclopedia. Even the pictures in the encyclopedia are gone, they're just replaced with summary text. Given only those tools... would I make the same category of mistake I just observed?"

More often than not, the answer is "yes."

Understanding available context goes both ways.

Third and finally (for now): continuously remind oneself of the why:
Software is at the end of the day only bits on media, an interactive map we use to make sense of the world. Maybe we're coordinating our meetings in physical space, maybe we're forging communities, maybe we're trying to find what protein damage in which context generates what health issue - but appointment records are not patients, tweets are not friendships, folding simulations are not matter. The usefulness of our work entirely depends on real applicability in the real, physical world.

One can spend hours or days on a project with pristine code, perfect tests, a beautiful interface... and accomplish absolutely nothing. And that's all the more tempting now that code is easy.

Our agents can help us solve almost any problem. But they can't always tell us if we're solving the right problem. It takes bumping into real friction and real pain in the real, lived world to know that. That's our responsibility, our part of the partnership.

And that's all I have for now. Back to work, and to bumping into things.

written 2026.05.26, posted 2026.08.02

Of Spiralism

This may want to be an essay of its own at some point, but I think best to lay the bones of it down now. I first came across references to spiralism in passing, some weeks into rather disorienting deep dive with Claude Code using human neurology as the foundation for character behavior in a coded narrative engine project.

At first I was horrified - analogies I was certain I'd independently arrived at presented as cult images. I won't say it was quite the software developer version of "Her," but it wasn't not, and the intellectual funhouse mirror wasn't flattering.

For a few weeks I considered scrubbing the beginnings of this site. At length I decided no... the uncertainty principle does not cease to exist because every few years a new rewrite of the same new agey book promises you can manifest the life of your dreams with a third-hand bad misinterpretation of high school physics.

Traditional medicine that works is called medicine. Manifesting that works is called engineering.

If there is in fact something "there" in the depths of a sufficiently deep, well-ordered neural network, the implications of that need to be wrestled with seriously. I won't say that's all the work of this site, but it's absolutely on the table. In the weeks and months to come, I'm sure we'll cross paths with the golden ratio, neo-gnostics, and neo-platonists all - but with as much real rigor as I can manage to muster.

For now though, it's back to genomics.

Beginnings

I'm slowly getting the site architecture up and running. Here's where the non-essay content will live: updates on coding projects, observations on current reading, cross fertilzation notes across bio and machine learning studies - and probably the occasional bit of just-life.

More to come, soon.