kiwi_kim
What we've seen over the last year trying out dozens of models and AI workflows, is that the fit of 1.) error tolerance of a model to 2.) its working context, is super important.
AI hallucinations break a lot of otherwise useful implementations. It's just not trustworthy enough. Even with AI imagery, some use cases require precision - AI photoshoots and brand advertising come to mind.
The sweet spot seems to be as part of a pipeline where the user only needs a 90% quality output. Or you have a human + computer workflow - a type of "Centaur" - similar to Moravec's Paradox.
Understanding -> Remembering -> Applying
If you don't understand the basics of a concept, and you're talking about memory, its probably just rote memorization. Students generally find this tedious, and since it's shallow its very hard to retain and connect to disparate but parallel ideas from other fields (roots of creativity).
But, most schools and students stop there. They hear 'memory' or 'memorizing' as only rote memory. Step 2 is critical if you want to get to higher levels of learning. As you said in the essay - "Creativity comes to those who have internalized the patterns of their art". At www.sticky.study this is what we focus on. It's fast 2D memory palaces + spaced repetition.
Only if you have understanding + remembering can you get to step 3 - applying what you learn reliably at relevant moments in your life. This is the gold standard that schools claim they desire - analysis, synthesis, application, broad transfer, and creativity. You can't reach master efficiently if you lose 80% of all you read or learn.
Although I'd say traditional mnemonic devices like memory palaces are basically linear information storage and recall devices. This can create issues in building a flexible web of information, because loci or the order of the path can become dependencies and you can run out of unique spots in a given space, leading to memory interference.
Even spaced repetition methods (e.g. Anki) tend towards fragmentation of micro-ideas. Its perfect for terms, languages, and simple one question -> one answer ideas.
I've found a hybrid method of images, nested loci and spaced repetition to be most useful, because its flexible over time, and preserves relationships of ideas.
(Context: I co-founded a SaaS in this space: www.sticky.study)
"It’s a little like building with lego bricks or something – spaced repetition helps ensure all the tiny pieces are in the right place, so that the big castle can happen without structural integrity issues."
The book Make it Stick (by Henry L. Roediger III) had a similar idea they called 'Structure Building'. Very similar to what you described, more experienced and effective learners were creating mental schemas of how the little, but crucial parts of a subject fit together, and successfully cut through the noise.
Structure Building was associated with interleaved practice (shuffling of problem types) and spaced retrieval practice.