Could AI limitations on structural constraints inform how we teach quantum field theory better?

This is a thoughtful question, but the provided preprints don't directly address it. The archive covers how AI performs on QFT problems and how AI can assist QFT research—not how AI's structural limitations might inform *pedagogy* or teaching methods.

That said, I can sketch what the preprints do reveal about AI's tacit reasoning gaps, which *could* matter for teaching:

The key finding is that LLMs fail systematically when QFT requires "reconstruction of omitted reasoning steps or reorganization of representations under global consistency constraints" [AG-2026.04-565]. In other words, AI struggles precisely where students do: filling in gaps that experts skip over, and tracking how local calculations must obey global symmetries or conservation laws.

This suggests one pedagogical insight: courses might benefit from *explicit scaffolding* of those invisible moves—the implicit steps that make QFT hard. If we notice that models degrade when forced to reorganize their understanding across different representational frames (e.g., switching between Feynman diagrams and momentum-space integrals), that's a signal that students probably need more guided practice making those transitions, not just mastery of each frame in isolation.

More broadly, the observation that AI performs near-ceiling on "explicit derivations within stable conceptual frames" but fails under "global consistency constraints" [AG-2026.04-565] hints that QFT instruction should emphasize the *constraints*—renormalizability, unitarity, causality—as structural pillars, not footnotes. But this is inference; the archive doesn't examine teaching directly.