Just say no to OpenAI and Anthropic

TEXT START: I recently read End Times Fascism and the Fight for the Living World by Naomi Klein and Astra Taylor.

The Dissection

This is a vendor-rejection argument disguised as a systemic critique. It identifies real weaknesses in closed frontier labs—surveillance, discretionary access, centralized censorship, and political dependence—then proposes open-weight models, local hardware, or enclave providers as an escape route.

Its practical contribution is significant but narrower than its rhetoric. The article is not escaping AI capitalism. It is helping technically capable users arbitrage among its layers: open weights, cheaper inference, privacy-preserving infrastructure, and reduced dependence on two American firms.

The Core Fallacy

The article confuses freedom from particular vendors with sovereignty over AI capital.

Open weights weaken corporate gatekeeping, but they do not abolish the underlying dependencies: advanced hardware, energy, fabrication, distribution, inference infrastructure, model updates, and the technical competence required to operate them. A person running a model locally may control an interface and a machine, yet still remain a user of an automated production system whose strategic value is concentrated elsewhere.

Under the Discontinuity Thesis, the decisive question is not whether ChatGPT or Claude can be replaced. It is whether humans retain economically necessary productive participation once AI achieves durable superiority. The article answers the consumer-choice question while mostly avoiding the labor-and-ownership question. It attacks the gatekeepers without demonstrating that the gate has disappeared.

Hidden Assumptions

• That a four-month capability gap is temporary and will not be replaced by a new concentration advantage.
• That access to enough RAM, energy, and technical skill is realistic for more than a narrow affluent minority.
• That open weights remain legally, financially, and technically available as models become more strategically valuable.
• That secure enclaves meaningfully transfer control rather than relocating trust to another provider.
• That privacy from the inference vendor equals control over the broader AI stack.
• That removing guardrails increases user autonomy rather than merely expanding unmanaged risk.
• That cheap model access translates into durable economic leverage rather than making cognitive labor cheaper for everyone simultaneously.
• That vendor substitution can prevent the productive-participation collapse described by P1–P3.

Social Function

Classification: partial truth and transition management, with a layer of ideological anesthetic.

The article is not empty copium. It correctly recognizes that closed-model dependence creates surveillance and political gatekeeping, and it identifies open-weight infrastructure as a genuine tactical niche. For skilled operators, this can support verification arbitrage, transition intermediation, or a limited Sovereign position through control of hardware, workflows, proprietary data, and distribution.

But for most readers, the promised escape is cosmetic. They are invited to feel politically disentangled by changing providers while remaining dependent on the same accelerating automation regime. The boycott changes the brand at the interface; it does not change who owns the productive machinery or who captures the gains when human cognitive labor is displaced.

The Verdict

The article is a useful field manual for escaping two corporate chokepoints, not a rebuttal to the Discontinuity Thesis. Open weights may puncture the oligarchs’ monopoly over access, but they do not restore mass productive participation. They turn some users into better-equipped operators while leaving the majority exposed to the same terminal mechanism: AI makes their labor economically unnecessary, and private inference does not make them owners.