tensorgami

The prevailing eschatology of the artificial intelligence boom has long been dominated by the binary of the Singularity: either the machine becomes a god, rendering all human labor economically valueless, or it remains a mere tool, subservient to human command. This dichotomy, however, has obscured the far more specific and brutal reality emerging in the mid-2020s. We have not achieved Artificial General Intelligence (AGI), with its implicit promise of universal capability and causal reasoning. Instead, as the mathematician Terence Tao has astutely observed, we have arrived at “Artificial General Cleverness” (AGC). While AGI would act as a universal leveler, AGC operates as a ruthless high-pass filter for the human intellect. It does not destroy work indiscriminately; rather, it targets a specific stratum of the socioeconomic hierarchy – the credentialed, managerial middle – by commoditizing the very asset that sustained it: mimetic competence.

The distinction Tao draws between intelligence and cleverness is not merely semantic; it is the boundary condition of the new economic reality. Intelligence implies a ground-truth understanding, a capability to derive conclusions from first principles and navigate the “jagged frontier” of novel problems with reliability. Cleverness, conversely, is the ability to recognize patterns, regurgitate heuristics, and simulate competence within established protocols. For the last half-century, the professional-managerial class thrived on Crystallized Intelligence () – the accumulation of facts, bureaucratic procedures, and social shibboleths. The value of the “midwit” – a term here used structurally to denote the median expert – lay in their ability to act as a friction layer for information, formatting the chaotic inputs of reality into the polished outputs of the corporation.

The arrival of AGC destroys the scarcity of this polishing mechanism. Large Language Models have proven themselves to be the ultimate mimics, capable of generating strategy documents, compliance reports, and boilerplate code that are aesthetically indistinguishable from, and often superior to, human output. When a machine can perform “performative competence” instantly and at negligible marginal cost, the human career built on performative competence evaporates. The danger for this class is not merely that the AI is “smarter” than them; it is that the AI exposes an ontological humiliation. It reveals that their professional contributions were never “knowledge work” in the generative sense, but merely stochastic token prediction performed by biological hardware – a process now infinitely optimized by silicon.

This displacement is exacerbated by the phenomenon Andrej Karpathy has termed “Vibe Coding,” which creates a perilous illusion of leverage. By allowing operators to dictate the “vibe” or intent of a software system while the AI handles the implementation, we have seemingly lowered the barrier to entry for technical creation. Yet, as Ilya Sutskever noted with his critique of the “jaggedness” of AI capabilities, this abstraction is a trap. The AI can bridge the gap between intent and execution only so long as the problem resides within the manifold of its training distribution. When the system drifts into edge cases – as all complex systems inevitably do – the abstraction collapses. For the midwit, who lacks the technical depth to descend from the abstraction to the code, “jaggedness” is not merely a bug; it is a risk of ruin.

Here lies the “Verification Saddle,” the mechanism that will drive the labor market bifurcations of the coming years. The shift from writing code (or policy) to prompting it is a shift from production to verification. Production requires only that one knows how to follow the steps; verification requires that one understands why the steps exist. The “Elite” architect, possessing high Fluid Intelligence (), uses AGC as a force multiplier. They can inspect the stochastic output of the model, identify the subtle hallucination in the logic, and correct it, thereby achieving a productivity multiplier of . Conversely, the manager or developer possessing only average fluid intelligence () is trapped. Lacking the depth to verify the output from first principles, they are forced to trust the “clever” machine. When the model inevitably hallucinates, this operator propagates the error, transforming from a producer of value into a vector of liability.

Consequently, we are witnessing the insolvency of “mimetic capital.” For decades, the university system and corporate ladder rewarded the ability to signal intelligence through the mastery of form – the correct jargon, the proper citation, the polished presentation. This corresponds to the “Wadsworth Constant” of human capital: the bloated signaling layer that surrounds the kernel of actual cognitive utility. AGC strips this layer away. The modern university, serving as the primary vendor of these rapidly depreciating signals, faces a liquidity crisis of relevance; it continues to sell certificates of in a market where the marginal cost of has fallen to zero.

Crucially, this collapse is not merely an internal accounting correction; it is a public spectacle. The market ceases to pay for the appearance of competence, and the general public ceases to believe in it. When a single architect using AI outperforms a government bureau or a legacy corporate department, the illusion of administrative complexity – the shield behind which the midwit has hidden – shatters in plain view. The correlation between raw cognitive horsepower and systemic stability becomes impossible to ignore, forcing a transparency that the managerial class cannot survive. The “Extinction Burst” predicted for the near future will not be a battle of man versus machine, but a frantic attempt by the intermediate cognitive strata to obscure the glaring reality of their own obsolescence. We are moving from an economy of production to an economy of judgment, and judgment, unlike procedure, cannot be memorized.