Fallacies¶
Incorrect beliefs about AI tools that reliably produce poor outcomes.
Fallacies differ from anti-patterns: an anti-pattern is a wrong practice, a fallacy is a wrong mental model. The same wrong belief tends to generate many wrong practices.
Pages¶
- The AI Knowledge Generation Fallacy — the belief that AI generates genuinely new information, and why it leads to misuse in architecture and test design
- Chain-of-Thought Reasoning Fallacy — visible reasoning steps are generated after the answer, not before; coherent explanation does not mean correct decision
- The Consistent Capability Fallacy — the belief that success on one task predicts success on similar tasks, and why capability doesn't generalize
- The Effortless AI Fallacy — the belief that AI should work without effort, and why it produces the worst results
- The LLM Laziness Deficit Fallacy — the belief that agents can be instructed into laziness; restraint must be enforced by harness gates, not prompts
- The Synthetic Ground Truth Fallacy — the belief that AI-generated artifacts can substitute for human-verified ones, and why it introduces compounding feedback loops
- The Task Framing Irrelevance Fallacy — the belief that surface framing doesn't matter, only the underlying problem does