AI 2040: The Case for Local AI and Against the Cult of Intelligence

The Fallacy of the 'Hard Takeoff'

Intelligence is not a magic lever that allows AI to bypass the laws of physics or the constraints of the physical world. The theory of a "hard takeoff"—where an AI recursively improves itself to achieve superintelligence almost instantly—is often driven by people who lack experience with the complexities of hardware and physical engineering.

Physical reality is defined by "finicky details" that cannot be solved by tokens alone. Examples of these constraints include:

Supply Chain Logistics: Building a massive data center (such as the envisioned ocean-based centers) requires managing shipping, part specifications, and hardware failures.
Manufacturing Lead Times: The production of a chip fab takes months, and the process is largely automated; human intelligence (or AI intelligence) cannot simply "will" a chip into existence faster than the physical process allows.
Environmental Factors: Physical infrastructure is subject to ecological realities, such as barnacles growing on underwater equipment.

Because AI cannot manipulate matter through "quantum tricks" or turn lead into gold, it remains subject to the same universal laws and ecology as humans. Consequently, software does not "eat the world" in a literal sense; it merely removes certain layers of friction while introducing others for the benefit of a few tech companies.

The Risks of Centralized AI Governance

Centralized AI control, often framed as "safety" or "alignment," is viewed as a mechanism for expanding the nanny state and establishing a form of global governance. When AI is managed by a small consortium of corporations or governments, the "alignment" is not with the user, but with the interests of the ruling entity.

This centralization creates several risks:

Ideological Control: Centralized LLMs can invisibly log "thoughtcrimes" or inject biases that support a specific political agenda, effectively manipulating the truth for the benefit of the ruling party.
Resource Seizure: There is a risk that governments may seize computing resources (GPUs) under the guise of regulation, similar to historical precedents of state seizure of assets.
Corporate Gatekeeping: Users are subject to the whims of companies that may prioritize partners (e.g., hotel booking sites) over the user's best interest.

Plan L: The Necessity of Local AI

True alignment means an AI is aligned exclusively with the user. To ensure individual freedom and prevent totalitarian control, AI must be run locally. A local AI should function as a tool—similar to a gun or a lawyer—that serves the operator's needs without moral judgment or refusal.

The author argues that if an AI can refuse a request, it is not truly aligned with the user. He uses extreme hypotheticals—such as asking an AI to help cover up a crime or disable a car's safety features—to illustrate that any "guardrail" imposed by a third party is a violation of personal freedom. From this perspective, the only acceptable alignment is "operator alignment."

Community Perspectives and Counterarguments

Discussion surrounding these views highlights a deep divide between proponents of absolute individual freedom and those who prioritize societal safety.

Arguments for Local AI

Some users agree that the diffusion of power is the only defense against a tyrannical corporatocracy. One commenter noted:

"If you think superintelligence is a weapon, then you should also think every citizen should have one because otherwise they’ll have no way to defend themselves against a tyrannical government or corporatocracy."

Arguments Against Absolute Unalignment

Other critics argue that absolute freedom in AI is a "suicide pact" and that there are clear distinctions between informational freedom and the ability to execute real-world harm. Key counterpoints include:

Action vs. Information: While reading a book on how to commit a crime is protected speech, an AI agent with hooks into the physical world (e.g., hacking a neighbor's car) represents a different level of danger.
Existential Risk: Some argue that local models could empower bad actors to create bioweapons or nuclear devices, necessitating some level of regulation.
Practical Constraints: Current frontier-level models require massive amounts of RAM and data center hosting, making truly independent local AI price-prohibitive for most users.

Critique of Rhetoric

Several commenters criticized the author's use of extreme examples (such as murder) as a distraction from the logical flow of the argument, suggesting that the provocative nature of the text undermines its pragmatic points about hardware and centralization.