I think this is pretty insightful, and I might add this as another reason LLM code looks so revolting. It's basically writing prose in a different language, which make sense - it's a _language_ model, it has no structural comprehension to speak of.

Whereas I write code (and expect good code to be written) such that most information is represented structurally: in types, truth tables, shape of interfaces and control flow, etc.

> I think this may be related to how people read code. You have people who scan shapes, and then you have people who read code almost like prose.

I think this is an astute observation.

I think there is another category of "reading" that happens, is what you're reading for "interaction" or "isolation".

Sure c.method is a scalable shape but if your system deals with Cats, Camels, Cars, and Crabs that same c.method when dealing with an abstract api call divorced from the underlying representation might not be as helpful.

I would think that we would have more and better research on this, but the only paper I could find was this: https://arxiv.org/pdf/2110.00785 its a meta analysis of 57 other papers, a decent primer but nothing ground breaking here.

> I scan shapes. ... verbal description.

I would be curious if you frequently use a debugger? Because I tend to find the latter style much more useful (descriptive) in that context.