Large language models are increasingly marketed as possessing "intelligence" through chain-of-thought reasoning, but critics argue this concept is a scientifically unsupported myth used to drive commercial dependency.

Key Points

• Proponents claim that increasing "reasoning effort" and token usage in models like OpenAI’s o1 or Anthropic’s Claude enhances their intelligence.
• Research, including a 2026 study on reasoning models, suggests that chain-of-thought performance often collapses on complex tasks and relies on memorization.
• Statistical analysis shows a negligible correlation between the length of a model's internal "reasoning" tokens and the quality of its final output.
• Critics argue that the reification of "intelligence" in AI is a marketing tactic designed to commodify cognitive labor and foster user addiction.
• Historical research indicates that "intelligence" lacks a consistent definition and has frequently been used to justify ideological or discriminatory agendas.

Why it Matters

The framing of AI as an "intelligent" utility encourages users to outsource critical thinking and accept opaque, potentially unfaithful reasoning processes as objective truth. By treating intelligence as a sellable product, tech companies risk creating a cycle of dependency that prioritizes corporate metrics over actual model reliability or transparency.