Focus on People, Models, and Thought
This note was prompted by Erik Larson's criticism of overconfident AI narratives.
The useful point is that intelligence cannot be reduced to impressive output alone. People carry models of the world, social context, physical expectations, goals, memories, and interpretive habits that are not captured by a surface transcript.
Machine-learning systems should therefore be evaluated by what kind of model they build, what they can test against the world, what they cannot know from text alone, and how their failures appear when the situation changes.
That keeps the discussion grounded: language models, quantum interpretations, consciousness claims, and theories of mind are not interchangeable. Each has to say what mechanism is being proposed and what evidence would distinguish it from a good imitation.