On what level of topics do you query LLMs to help you learn? As you become a domain expert, the usefulness of LLMs diminishes, I imagine?
Regarding the learning of learning from first principles vs from 'working interfaces', could this be said as 'learning top to bottom' (high-level ideas first) vs 'bottom to top' (low-level first, e.g. understanding axioms of real numbers before understanding algebra)?
It doesn't diminish, the knowledge is like a fractal. You can zoom out, in, etc.
I created a prompt that asks me a set of questions on a topic. It then scores my understanding, gives me deeper insight into the topics, broadens my understanding with some extra information, and then provides a mind-map of related topics, and a mind-map of adjacent topics.
To bootstrap an area of understanding, I ask it for a mind-map of a topic. Like a fractal, you can choose a line item and go deeper.
Prompt 1: Give me a small mind map on the amygdala
Based on that, prompt 2: give me a mind map on the role of the basolateral nuclei of the amygdala in the creation of phobias
I find that domain expertise is really a funny mirage in a lot of ways.
What I want in reading a book is less often book-domain expertise and more often "convert book's contents to my thinking style and existing systems fit." Then I will typically use that result in the domains in which I'm already an expert...that's what I generally find that I want.
For this reason the LLM is helpful in abstraction duties. It doesn't by itself need to be as much of a domain expert, as much as it needs to give me e.g. more fluid interface-mindset access.
On the other hand, if it is lacking even basic quality of internal schema, sure? It's just that this hasn't happened to me yet; maybe that's surprising but it's more likely that this happens based on my own specific questions and theories than with the contents of published books.
So many of the books are functionally general in nature but still foreign in their nomenclature, etc. and I find that this is a good leverage point for an LLM.
> could this be said as
It usually is said that way, in my experience. I personally find it more helpful to talk about interfaces vs. internals. But others may be more used to talking about top-down.
For example with the interfaces metaphor, you get to have more than one "top" which maps very well to systems focus & benefits IMO. The leverage here is very useful in some teaching and learning cases.
Regarding the learning of learning from first principles vs from 'working interfaces', could this be said as 'learning top to bottom' (high-level ideas first) vs 'bottom to top' (low-level first, e.g. understanding axioms of real numbers before understanding algebra)?