Deepen lecture 14: flip lens to examine human cognition limits

Major revision focusing on what LLMs reveal about human understanding:

Content changes:
- New framing: models as mirrors for human cognition, not just evaluation targets
- Discussion prompts now explore human cognitive limits, not just LLM capabilities
- Added 'spectrum of grounding' discussion (coffee → quark → GPT)
- Added 'boundaries of human understanding' section (cognitive closure)
- Added 'illusion of explanatory depth' (Rozenblit & Keil)
- Removed shallow examples (coffee sensory list, simple Word2Vec failures)
- Deeper questions: Are humans stochastic parrots too?

Style changes:
- All titles converted to sentence case
- Scale classes added for overflow slides
- More concise: 491 lines (was 684)
- Removed redundant content covered in prior lectures

New discussion questions:
1. Where do you draw the line on 'real' understanding?
2. What concepts do you suspect you fundamentally cannot grasp?
3. Is understanding a spectrum rather than binary?

Author: jeremymanning