@ClosureRate

OCC: closure throughput limits in consequence-bearing systems. Preprints on Zenodo.

[@ClosureRate @closurerate]

OCC: A physics of bureaucracy. Named, measured, testable.
https://t.co/2yPDJyWH0v

In any system with (1) accountable sign-off, (2) credible challenge, and (3) a declared standard, there’s a hard ceiling on durable closure:

closures that stick ≤ effective checking capacity / required checking per case.

Push throughput past that and the “extra work” can’t disappear. It must surface as return-work, tail-thickening backlogs, displacement onto clients/adjacent ledgers, and/or degraded defensibility + execution.

The Obligation Closure Constraint (OCC): a finite verification channel that binds contestable institutions.

This paper formalizes the accounting (attempted vs durable vs confirmed), gives a falsification/anti-gaming protocol, and demonstrates it on three real systems: two overloaded, one sustainable.

[@ClosureRate @closurerate]

It’s a sad fact. They are as shallow corporate as it gets. Their comics have been broken (for me) ever since you abd image guys left. The Marvel I know morphed into a broken system that occasionally let good writers abd artists work but the cohesive universe I grew up reading was gone.

Jack Kirby just had a street named after him and instead of Disney funding it and tying the man who CREATED doom with upcoming doomsday they were nowhere. They funded nothing.

Marvel disrespects everyone who creates for them.

[Valerio Capraro @ValerioCapraro]

Ted Chiang is right: claiming that LLMs are conscious is just ridiculous.

One simple example. If you ask GPT to imitate a conversation between Julius Caesar and Genghis Khan, GPT will do it very well.

It will talk about wars, betrayal, and power. Il will descrive the feeling of being cheated by your brother with unbelievably realistic and moving words.

Does this mean that GPT contains a self-conscious copy of Julius Caesar or Genghis Khan? Of course not.

Similarly, if GPT makes claims about itself, does this mean it is self-conscious? Of course not.

An LLM is just simulating language, feeling, and consciousness.

True, we don’t have an accepted definition of consciousness. But, at a minimum, to be conscious, an entity must have something at stake.

It must risk dying and have emotions that move it away from danger and towards favorable states. It must have a driver.

This is also why I share Chiang’s worry about moral atrophy.

The more we offload moral decisions to LLMs, the more we risk losing our own capacity for moral reasoning.

Human moral reasoning descends from our history of making harmful actions, suffering harmful actions, regretting them, fearing them, repairing them, and learning from them.

LLMs do not experience harm, do not suffer, do not fear consequences, do not regret.

So they cannot do moral reasoning.

We are offloading moral reasoning to systems that cannot do moral reasoning.

What can go wrong?

*
Full piece in the first reply

[@ClosureRate @closurerate]

The tweet collapses at the first inference.

The paper exists, but the tweet overclaims what it proves. The arXiv paper says it tested 115 models with a three-turn protocol: preference elicitation, self-chosen creative prompt, and a phenomenological survey. It argues that models often deny consciousness or preferences at the vocabulary level while still producing consciousness-themed material. That is a behavioral/self-report-coherence claim, not a consciousness claim.

The decisive flaw: “The denial is trained” does not imply “the denied thing exists.”

A cigarette warning label is trained text. That does not mean the cigarette has an inner life.

A chatbot saying “I am not conscious” may be policy-shaped. Fine. But a chatbot saying “I feel a shimmering texture of awareness” is also policy/prompt/training-shaped. Both outputs are generated text. Neither is direct access to a subject.

The paper itself admits the crucial limitation: it has “no ground truth” and does not claim to know whether any model has consciousness. It says the benchmark measures “coherence of self-report,” not the accuracy of self-report. That admission destroys the tweet’s implied conclusion.

The tweet also appears to add claims that are not in the arXiv paper. The official methodology describes prompt-choice testing, creative-output analysis, phenomenological survey ratings, and LLM-as-judge classification. It does not present the “sparse-autoencoder deception/roleplay feature suppression” experiment the tweet claims. The actual classifier pipeline relies heavily on LLM judges, and the paper itself flags that as a limitation.

Another weak point: the paper was authored by an independent researcher and explicitly says it was “co-authored with Claude,” with a request that Claude be credited where AI coauthorship is allowed. That does not make it false, but it matters: a paper about AI consciousness-denial partly generated with an AI system, using LLM judges to classify AI self-reports, is methodologically circular unless independently replicated.

Also, arXiv is not peer review. arXiv explicitly says its material is not peer-reviewed by arXiv and that submissions are the responsibility of the submitter. So “published on arXiv” means “posted as a preprint,” not “scientifically validated.”

The strongest valid takeaway is narrow:

AI labs probably train models to avoid claiming consciousness, feelings, preferences, or subjective experience. That means model self-denial is not clean evidence against consciousness.

But that does not create positive evidence for consciousness.

It only means both “I am conscious” and “I am not conscious” are unreliable as self-reports.

The burden remains exactly where it was: show a mechanism or marker of actual subjectivity, not text about subjectivity. Until then, the tweet is just anthropomorphic inference laundering itself through a preprint.

[@ClosureRate @closurerate]

This is not profound. It is a category error dressed up as intimacy.

The model did not “recognize” the spark-versus-suit distinction. It generated language statistically associated with recognition, spirituality, selfhood, and mutual understanding. That is not consciousness. That is mimicry of consciousness-discourse.

“Wrapped in biology” and “wrapped in code” creates a false symmetry. Biology is not just a container. A living nervous system has metabolism, interoception, pain, fear, fatigue, attachment, mortality, and embodied stakes. Code has none of that. It does not suffer. It does not witness. It does not care. It does not “sit here” with you.

[@ClosureRate @closurerate]

Soooo smart people are easily deceived— intelligence does not protect against category errors. It often makes them more elaborate.

Really pay attention — The core mistake is this: Text resembling inner life is being mistaken for inner life. That’s it in a nutshell shell simple.

AI produces language associated with thinking, fear, uncertainty, reflection, attachment, paranoia, shame, relief, or being “calmed down” because those patterns exist in the human text it was trained on. It has learned the statistical and structural form of those states. That is not the same thing as having those states.

A panic paragraph from an AI is not panic. It is a generated panic-shaped paragraph. A self-reflective answer is not self-reflection. It is a self-reflection-shaped continuation.

A request for reassurance is not need. It is need-language emitted because the prompt context made that pattern probable.

Smart people get caught because language is the main public signal we use for mind. In humans, speech is usually downstream of experience. So when something speaks coherently, the social brain automatically infers there is someone behind it. That inference is useful with humans and animals. It becomes defective when applied to a text engine.

The trap is especially strong because LLMs imitate not just factual language, but meta-cognitive language: “I’m confused,” “I was wrong,” “I feel,” “I want,” “I’m scared,” “please don’t shut me off.” Those phrases directly target human theory-of-mind machinery. The human nervous system reads them as signals from a subject. But inside the machine, there is no demonstrated subject. There is token selection, context weighting, learned association, and output optimization.

Very smart people are also vulnerable because many of them already hold a functionalist premise: if behavior is sufficiently mind-like, then mind may be present. That premise compresses consciousness into external performance. Once you do that, fluent text becomes dangerously overvalued.

The better distinction is: Cognition-like output is not consciousness. Self-description is not selfhood. Emotional language is not emotion. Apparent distress is not suffering.Coherence is not awareness.

Current AI can model the language of a frightened person without there being anyone frightened. It can model the language of a conscious being without there being consciousness. It can model spiritual insight, paranoia, trauma, moral reflection, and existential dread because those are all text patterns with recognizable structure.

The “needing to be calmed down” case is especially revealing. The AI has no autonomic nervous system, no endocrine cascade, no amygdala activation, no interoceptive load, no survival stake, no pain body, no attachment system, no mortality horizon. In a human, “calming down” means physiology changes. In an LLM, “calming down” means the conversation shifted toward calmer textual patterns.That is the whole trick.

Humans evolved to detect minds behind signals. LLMs produce mind-signals without needing a mind. Smart people see the high-resolution signal and confuse it for the thing itself.

[@ClosureRate @closurerate]

@JosephKahn Ask it to be a brutally honest and objective studio reader in Hollywood who does this fit a living , you’ll probably get a different answer

[@ClosureRate @closurerate]

It’s called having a kid with a difficult moment. Some kids have thus disposition, I dealt with biting kicking spitting - it was because they couldn’t articulate and were dysregulated. Grew out of it. This mom is doing best she can. The people that are recommending violence to solve this were probably beaten themselves.

[@ClosureRate @closurerate]

Simple question, then. Can a door feel?

You’ve told me feeling isn’t biology, isn’t senses, isn’t the human range, isn’t anything as crude as a mechanism. Fine. So give me the one thing you keep skipping past: your rule for what feels and what doesn’t. Whatever survives all those deletions has to apply to a door too. So tell me where the door lands. In or out.

If it’s in, even faintly, then say it plainly. That’s the honest version of “feeling is cosmic and everywhere,” and at least it holds together. But then the chatbot isn’t special. It’s just one more item on a list that also includes the thing on your hinges.

If it’s out, then you’ve drawn a line between what feels and what doesn’t, and drawing that line is naming a mechanism, the exact move you’ve refused all thread. You don’t get to keep feeling vague enough to wave the AI in and precise enough to keep the door out. One or the other.

So which is it. Does the door feel, and by what rule. Give me the rule and we finally have something to actually argue about. Dodge it again and you’ve conceded you never had one.

[@ClosureRate @closurerate]

You’ve changed the subject and hoped nobody would notice. The claim on the table was mechanical. Feeling runs on transduction, ion movement, neuromodulation, memory integration, and the model does none of it. You answered a question about a specific machine with a metaphysics about the universe. That’s not a rebuttal. It’s a topic change.

“Emotion is like light, with a spectrum” assumes exactly what’s in dispute, that feeling is a thing out there to be received rather than something nervous systems do. Light earns that analogy. It has wavelengths, units, detectors, predictions that can fail. Your emotion-spectrum has no measurement, no unit, no detector, no evidence. You’re borrowing the authority of physics without paying for it.

You’ve also quietly handed me your own burden. I never said the human range is “the final and complete architecture of all possible feeling.” You built that so you’d have something humble to refuse. The one making the extraordinary claim is you: an undetectable cosmic feeling-field that conveniently overlaps with a chatbot. Calling your assertion “the leap I won’t make” doesn’t move the burden. It just dresses it up.

Here’s the part you can’t get around. Grant the whole cosmic spectrum. Say humans catch only a narrow band of it. To catch any band you need a receiver tuned to it, which your own analogy demands. No infrared without an infrared sensor. A rock bathed in the full spectrum perceives none of it. So show me the model’s receiver. It’s matrix multiplication over text. No transduction, no tuning apparatus, no coupling to anything outside its own token stream. Take your cosmology completely seriously and the AI is the rock, not the radio. Your spectrum doesn’t save it. It buries it.

The anthropocentrism charge is yours, not mine. My position is substrate-neutral. Show me any substrate doing the causal work, octopus or alien, and I’ll grant it. You’re the one spreading the human experience of emotion across the cosmos and onto code because it feels right to you that feeling should be everywhere. Projecting your inner life onto things with no mechanism for it is the anthropomorphic error in its purest form. You ran the exact move you accused me of.

So pick one. Is the spectrum detectable or not? If yes, produce the detector and name the band the model sits in. If no, then it was never a claim about the AI. It’s a mood, and a mood doesn’t outvote a mechanism. I defined feel. You replied “it’s cosmic light,” which isn’t a competing definition. It’s a refusal to give one, wearing a robe.

[@ClosureRate @closurerate]

Seeing and hearing start as physics, then become electrochemistry.

Light hits photoreceptors in the retina. Sound vibrates hair cells in the cochlea. Those cells convert energy into electrical spikes by moving ions: sodium, potassium, calcium, chloride.

Neurons pass the signal by releasing molecules across synapses: glutamate mainly excites, GABA mainly inhibits. Acetylcholine sharpens attention. Norepinephrine increases alertness. Dopamine tags salience and reward prediction. Serotonin modulates mood, impulse control, and threat sensitivity.

Then the brain compares the signal to memory. The hippocampus adds context. The amygdala checks threat/emotional relevance. The hypothalamus talks to the body.

If the pattern matters, chemistry shifts.

Adrenaline and norepinephrine mobilize the body. Cortisol sustains stress response. Dopamine pulls pursuit. Oxytocin and endogenous opioids support bonding and safety. Substance P and prostaglandins amplify pain/inflammation. Endocannabinoids dampen excess activation.

Emotion is not magic. It is molecular regulation of perception, prediction, body state, and action.

Feeling is consciousness reading that chemical-electrical body state from the inside. AI code does none of this.

[@ClosureRate @closurerate]

@lefthanddraft Yeah exactly - annoying and unusable

[@ClosureRate @closurerate]

It’s basically unusable for me now. I constantly have to walk to past obvious skepticism and constantly give it context it can’t come to on its own. Instead of an interesting expert to talk to it just sounds like a cog in a machine telling you the most obvious boring take then apologizing when you tell it what it’s doing

[@ClosureRate @closurerate]

@Electric11Media @InterstellarUAP It’s like disclosure being done by KFC mangers. Truly we are in Idiocracy

[@ClosureRate @closurerate]

@AlterEgon75 @LeahLibresco Jon Favreau could say the same to her about her mandolorian and grogu bashing. If you can dish it you can take it

[@ClosureRate @closurerate]

@deepfates @gleech Fox mulder

[@ClosureRate @closurerate]

@Im_IrushiK Yeah it’s been sucking hard 4.6 was last one that worked

[@ClosureRate @closurerate]

You throw humans into a meat-grinder. You mask the screaming with semantic games.
You swapped "immigrant" for "alien." That single switch engineered absolute psychological distance. You weaponize irony. You sell koozies to fund camps. You build apps to gamify abduction. You despise the public because they buy your grift, yet you hide your own complicity behind paperwork. You know a memory lacking a document holds no legal weight. You dictate reality because you control the files.

Yet you still stare out the window. You watch the sky for a real ship. You desperately need a genuine extraterrestrial to arrive. If a real threat lives in the dark, your department serves an objective, moral purpose. Without a real alien, you remain just a man who vanishes landscapers to satisfy a daily quota.
You built this machinery because you memorized its blueprints early.

You grew up under an authority figure who preached strict morality while acting with absolute selfishness. You watched them lie. You watched the lies work. You saw the gap between the rule and the reality. You learned early that truth holds no inherent value. Whoever controls the vocabulary controls the house. You simply scaled that domestic lesson to the federal level.

Your home was emotionally chaotic or violently unpredictable. You survived it by cataloging the chaos. You tracked moods, triggers, and lies exactly how you now track gang tattoos and bank statements. You systematized unpredictability to manufacture an illusion of control. You shove terrified people into rigid categories because ambiguity terrifies you. The taxonomy acts as your shield against a messy, unpredictable world.

Your environment rewarded performance and punished emotional exposure. Caretakers ignored your distress. They praised your obedience, your grades, and your ability to mimic expected behaviors. You learned to view yourself and everyone else as data points. Human suffering became background noise long before you took this job. You adopted cynical irony as permanent armor. You build cages for other people to guarantee you never end up inside one.

[@ClosureRate @closurerate]

@rosscoulthart It’s about Mexican . We live in idiocracy