Rethink Your Mental Model in the Age of Generative AI: A Triadic Framework for Human-AI Collaboration

Generative AI systems such as large language models exhibit jagged intelligence: they combine superhuman performance on some tasks with brittle, often opaque failures on others. Existing mental models, inherited from deterministic technologies and from human teamwork, mischaracterize these systems systematically. This theory-driven synthesis develops a conceptual _Triadic Framework_ for adaptive mental models in human-AI collaboration: First, the _System Layer_ synthesizes current evidence on probabilistic generation, opacity at scale, and rapid model drift, explaining why capability boundaries are uneven and moving. Second, the _Collaboration Layer_ analyzes how users configure prompting practices, division of labor, and handling preferences. Third, the _Metacognitive Layer_ examines how anthropomorphism, metaphors, and cognitive biases shape human interpretations of “intelligent” behavior. Building on this diagnosis, the paper proposes seven actionable practices that are designed to calibrate mental models, preserve human agency, and make hybrid intelligence more robust under jagged and shifting AI capabilities. Together, the paper reframes the challenge of generative AI from crafting better prompts to cultivating more adequate mental models of probabilistic counterparts.