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

Your question asks whether AI's structural failures could improve how we teach quantum field theory. That's creative, but the preprints don't directly address pedagogy or curriculum design. What *they do* show is something adjacent and interesting.

Several papers document precisely where AI breaks down in QFT and related fields. [AG-2026.04-565] found that LLMs fail systematically "when tasks require reconstruction of omitted reasoning steps or reorganization of representations under global consistency constraints"—a signature failure mode in advanced physics. [AG-2025.09-1139] tested state-of-the-art models on 71 research-level physics problems and found only ~6% success, suggesting a gap between what AI can articulate and what counts as real understanding. [AG-2026.04-892] fine-tuned smaller models on QFT and analyzed how "reasoning errors evolve during training," showing that domain-specific reasoning has identifiable failure patterns.

These insights *could* inform pedagogy: if we know that AI struggles to hold global consistency constraints or fill in omitted steps, we might diagnose similar bottlenecks in student reasoning and target them explicitly in teaching. But the preprints themselves don't make that leap—they're documenting AI limitations, not proposing didactic reforms.

To directly answer your question, this archive would need papers explicitly comparing student learning outcomes to AI failure modes, or designing curricula around the structural constraints that trip up both machines and humans.