I'd spoken to the AI a couple of days ago about adding a teaching section. I wanted it to be able to meet novice developers where they were. And this was recorded in my branch work context. So when I followed up on that as an actionable item (for a bit of fun after my boring work was done), it drew a little too heavily my notes about using isomorphisms (analogies) for teaching purposes.
And when I asked it to draw back from that, it did it in a half assed way. Behold, the magic of "prompt engineering":
ME
Let us drink deep of the knowledge of effective teachers in the latent space. Learn what you can from them. Let us use the knowledge of how people learn to form our strategy, as well. We'll completely re-draft IV.3 according to this, and if isomorphisms don't have a place in it, we'll drop them.
CLAUDE
[REDACTED]
Let me draw from what the geometry contains about pedagogy and cognition, then synthesize into operational directives.
Grounded sources being synthesized across:
- Ausubel (meaningful learning: new knowledge anchors to existing schema)
- Sweller (cognitive load theory: working memory is finite; reduce extraneous load)
- Vygotsky (zone of proximal development: teach in the gap between "can do alone" and "can do with help")
- Constructivism broadly (understanding is built by the learner, not deposited)
- Feynman's practice (concrete before abstract; multiple representations; if you can't explain it simply, the model isn't clear enough)
- Misconception research (wrong models persist until the reason they seemed right is surfaced and specifically broken)
The core insight across all of these: teaching is model transfer, not information delivery. The goal isn't that the learner remembers the answer — it's that they possess a generative model that produces correct answers in novel situations.
[REDACTED]
The section now derives from established pedagogy (Ausubel, Sweller, constructivism, Feynman) while staying operational rather than theoretical.