Audit lecture 26: verify refs, fix citations, add new slides

- Move citations from box titles to inline hyperlinks with source URLs
- Add transition slide "From deception to detection" between reward
  tampering and mechanistic interpretability sections
- Add monosemanticity definition-box (Bricken et al., 2023)
- Add Hinton Nobel Prize (Physics, 2024) to existential risk slide
- Fix Bartz v. Anthropic date (Aug 2025, not Sep), add Thomson Reuters
  v. ROSS case, update fair use ruling count (3 rulings: 2 for, 1 against)
- Add Challenger Gray & Christmas and BLS source links on employment slide
- Add strategic dishonesty citation (Motwani et al., 2024)
- Add source links for deepfakes slide (FCC, Storm-1516, ISD Global)
- Add EU AI Act link
- Replace "Discussion: your line in the sand" with two "What's next
  for LLMs?" slides covering scaling limits (energy, data, model
  collapse), efficiency (LoRA, MoE, distillation, nested learning),
  local models, privacy, and multimodal progress
- Fix slide scaling (scale-65, scale-75) for content-heavy slides

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: jeremymanning