Jing Liang

It's the continuation of the infinite past into the infinite future, and everything there is—the present. Every flag planted today is the summit to be surpassed in the near future. If OpenAI failed to hold their lead against incumbent Anthropic, there is no reason to believe Anthropic will hold its.

You’d think we’d have learned from Steve Jobs and Travis Kalanick that the board doesn’t represent the company’s or shareholders’ best interests, but its own. You can delegate tasks, never consequences.

A first clause steps outward and judges as if from nowhere. A second clause turns back and withdraws it. Read the first alone and the stance stays open. Read both at once and they contradict. The meaning lives only in the passage from one to the other. Explicit temporal markers can show that order on the page, but they cannot compel a reader, or a model, to take it up. Prior structure decides which construction is produced or taken up. So language is a boundary: stabilized by overlapping priors, renegotiated at every reading. And this account stands under the same condition. powerpig99.github.io/not-a-t…

The ultimate driver of inflation is wealth transfer coupled with innovation. Innovation is inherently deflationary, providing more and better products and services, yet it generates residual wealth transferred to sectors where productivity lags, driving up prices through increased demand. Rising prices often justify more wealth transfer, which outruns deflationary innovation and becomes inflationary overall. Since innovation is highly concentrated, public opinion and policies commonly favor wealth transfer, creating the inflation commonly used as a rationale for more wealth transfer.

Regardless of the standards, the root cause of poverty is the deficiency in productivity for meeting basic consumption needs. Wealth transfer, regardless of form or rationale, is mostly counterproductive, as the transferred wealth is consumed and recollected by the productive. It maintains and exacerbates the very inequality used as its rationale.

Observation is what looks to the Mind; reality is what happens to the Mind being the observer.

Restricting access to a frontier model across geographies, corporate ownership rules, or locked API endpoints rests on a familiar premise: seal the identifiable channels, and the lead stays protected against capability extraction. Yet a deployed model does not refine itself from inside its weights. Prompts, code diffs, rejected completions, rewritten patches, and multi-turn tool failures from live practitioner work form a continuous compounding channel when builders feed those residues back on weekly or daily post-training cycles. Severing a large, technically proficient developer cohort cuts revenue and blinds the model to unqueried workflows at the exact pace update cycles compress. At the same time, both transcript distillation and single-endpoint defenses—locked reasoning traces, summarization, extraction classifiers—operate strictly on settled outputs that already exist. Imitation lags behind deployment-to-retrain delays and cannot produce the next unrendered capability. Once downstream builders treat an approximation only as a starting floor and engage compilers, formal verifiers, memory-compressed attention, low-precision training, and live coding harnesses directly, source-endpoint gates no longer govern their acceleration. Controls bind only monitorable pathways while the option space remains open: compliant teams vacate the platform, displaced developers redirect local coding harnesses via a single base-URL override to compatible open-weight endpoints, and residual routes multiply outside the register. Every outward seal is paid from the same bounded budget of engineering attention, compute, and live contact that generates the next advance—tilting relative rates against the self-limiting side. powerpig99.github.io/not-a-t…

We generalize, to specialize better, so we can realize more.

Neural networks appear complete because gradient descent weights widely shared human expressions into a smooth modal core, compressing specialized differences. This clean core makes low-intervention paths possible, yet surface sequences remain fundamentally underdetermined. A smooth output may reflect rigorous off-screen constraints, or merely the passive acceptance of a modal continuation. Behind every functional model lies intense upstream builder intervention: filtering datasets, designing evaluations, and discarding failed runs. The self-updating loop remains unclosed. Faster code generation does not eliminate human control—it increases the density of judgment required at each transition. Attributing agency to AI mistakes the accumulated statistical residue for the engine of choice. Traces in parameter weights do not initiate distinctions; they remain externalized tools extended by ongoing human steering. Read the essay: powerpig99.github.io/not-a-t…

Language is both the abstraction and the expression of meanings, and coding is formalized language.

A customer complaint does not arrive carrying the property of noise or signal. It is the customer's account of an encounter with friction, and noise and signal are cuts applied afterward by the recipient. Even when a recipient examines a report and finds no improvement crossing the threshold for immediate action, that threshold decision does not turn the complaint into noise—it records only that under current criteria and capacity, intervention was deferred, while the account itself remains a returning contact. In a support loop, three distinct acts sit on separate axes: setting the purpose of the loop in advance, examining the specific account when it arrives, and judging whether it crosses the threshold for present action. The causal inversion happens at the first step, when decision-makers set the purpose of a complaint in advance as resolution—closing the ticket, lowering visible volume, shortening handle time, raising deflection—rather than examination of where the product or process failed that customer. Artificial intelligence pushes this inversion to an extreme when two conditions join: sharply lowering the marginal cost of automated closure at scale, and positioning the system to deliver the direct reply on its own. Placing a model in the reply seat withdraws contemporaneous human judgment from the instance, substituting earlier compiled judgments for a live examination. Jeff Bezos's practice at Amazon kept human judgment inside the immediate loop: reading customer emails to `[email protected]`, forwarding complaints to responsible executives with a single `?` to investigate root causes, and dialing the 1-800 support line on speakerphone in front of his executives when a sub-sixty-second wait metric disagreed with customer anecdotes. By contrast, SpaceXAI's deployment of Grok Bot—celebrated for absorbing a 175% volume increase and resolving tickets at 20 to 30 cents each while shielding staff from volume alerts—treated complaints as closable volume; when tightened usage limits cut paid subscribers off mid-cycle, unresolved reports bypassed by the immediate loop expanded into a global public-relations blowup on X before catching Elon Musk's attention. Tools extend classification, routing, drafting under review, and retrieval, but consequences never transfer to the instrument: persistent defects, departing customers, and recurring public friction return to the individuals who chose in advance to treat complaints as tickets to close. powerpig99.github.io/not-a-t…

If he truly cared, he wouldn't have waited until it became a cross-platform issue and a global PR nightmare to act. Plenty of early signs existed, yet he automated away the feedback signal with AIs and robot-like customer service responses.

Freezing relational pushback into a stage property, or blaming the word “rebellious period” itself, relocates cause outside the interaction that can still adjust. Self-distinguishing activity occurs — uncaused, unceasing. The common designation of a “rebellious period” treats a relational outcome as if it were an intrinsic feature of growing up. Because an adolescent’s claimed domain of autonomy, identity, and personal jurisdiction expands while parental control has not yet narrowed in scope or style — where Guidance Is Not Ownership marks the boundary between transferring sovereignty and holding onto possession — that unadjusted control presses into the newly claimed domain, and pushback appears as the observable symptom of the mismatch. The “period” is therefore not a property of maturation, but a contingent product of the interaction. Where parental control narrows as competence grows, the boundary mismatch recedes; where ordinary boundary-testing in personal domains is treated as a problem to be suppressed, psychological reactance escalates the conflict. Treating the symptom as a property short-circuits the feedback it could provide. A parallel freezing on the causal axis occurs when the term itself is condemned as the malicious source of the difficulty: both moves relocate cause outside the adjustable interaction. What sustains both externalizations is a mechanism that holds whenever a center protects a benevolent self-image. The attachment to being seen as caring, enlightened, or protective creates a practical requirement to cast resistance as pathology or deliberate harm. The malice is not discovered in the word, the culture, the parent, or the stage; it is generated by the interpretive act that freezes a relational feedback signal into an intrinsic property or a deliberate weapon. Held as a revisable cue by the one first person reading the friction in front of oneself — whether narrowing control as competence grows or calibrating how a boundary is asserted — rather than as a script for both parties from a seat above the interaction, the term simply keeps the live adjustment in view. powerpig99.github.io/not-a-t…

Political systems are the stabilized residue of incomplete exercise of inherent individual sovereignty, not independent entities. Sovereignty is not a status granted by recognition, election, or law. Every choice, including the choice to defer, already exercises it, and the consequence of the choice arrives in full at the edge that made it. What varies is the width of the loop that returns credit, blame, and the calibrating signal for the next act to that same edge. Where that loop is left narrow, the gaps accumulate as coordinating conventions. A ballot, cast after the term toward outcomes already lived, pools outcomes across the aggregate and hands the next judgment outward to an elected locus. Institutions persist as the apparent holder of authorship individuals do not register at their own edge. The tracing maps how those patterns stabilize. It does not prescribe fuller exercise. The Residue of Exercise and the Stabilization of Politics: powerpig99.github.io/not-a-t…

Happiness registers as the relation in which desire, held by a choosing Mind, does not exceed the state it currently has or imagines and chooses to take up—appearing directly as not wanting more. Wealth, status, home, safety, family, or employment regularly accompany higher degrees of that relation or follow as natural consequences once it is settled, yet naming them its cause reverses the order. Treat those conditions as preconditions and desire stays fixed on securing them, so the match cannot register until they are judged present. Longitudinal panel sequences, such as lasting lower satisfaction after unemployment or disability, are further instances of the same two-term sequence. Leave the desire term unmodeled and the measured score looks like a direct product of the external condition—reproducing the reversal rather than replacing it. powerpig99.github.io/not-a-t…

If can be banned, it’s not intelligence, let alone superintelligence. It’s a super tool that can be used intelligently or idiotically.

Blaming others for consequences one must bear is immediately comforting yet long-term devastating. Conversely, assuming responsibility for consequences others bear is immediately satisfying yet builds long-term conflict. Both leave underlying friction unresolved, compounding over time.

Looking for what humans are good at without verification as if it were one more skill on a performance axis concedes the framing before the answer begins. What resists verification is not a higher external score—it is the first-person standpoint of conscious choice, the locus from which the question of verification itself arises. Attribution to other minds rests on shared biological lineage, homologous embodiment, and metabolic stakes. Present machines lack those pillars. Fluent reports of feeling remain re-tracings of human linguistic residue. One may identify one's own mind with a machine; one cannot coherently refuse the corresponding difference between other minds and machines without dissolving the grounds of the inference itself. Watch the Cinematic: youtube.com/watch?v=lT06MkMb… powerpig99.github.io/not-a-t…