How AI's Actually Work — HAAW

Index

The HAAW pages in reading order, from the physical AI machine through inference, training, agents, control, observability and mission reconstruction.
45 pages. Each full title is the link and opens in its own named target.
  1. 01 HAAW — The AI Machine: Before Anything Happens
  2. 02 HAAW — Machine Boots
  3. 03 HAAW — The Model Loads
  4. 04 HAAW — The Prompt Arrives
  5. 05 HAAW — Tokens Become Vectors
  6. 06 HAAW — Inside the Transformer
  7. 07 HAAW — Choosing the Next Token
  8. 08 HAAW — The Generation Loop
  9. 09 HAAW — The Answer Emerges
  10. 10 HAAW — What IS the LLM?
  11. 11 HAAW — Inside the GPU: Where the Math Runs
  12. 12 HAAW — Tensors and Matrix Multiplication
  13. 13 HAAW — Weights and Activations
  14. 14 HAAW — Memory in Motion
  15. 15 HAAW — More Than One GPU
  16. 16 HAAW — The Inference Runtime
  17. 17 HAAW — The Interface
  18. 18 HAAW — Context
  19. 19 HAAW — Training
  20. 20 HAAW — Backpropagation
  21. 21 HAAW — The Optimizer
  22. 22 HAAW — Training Data
  23. 23 HAAW — From Pretraining to Post-Training
  24. 24 HAAW — Evaluation
  25. 25 HAAW — How Does a Model Learn to Reason?
  26. 26 HAAW — Inside a Reasoning Step
  27. 27 HAAW — Interpretability
  28. 28 HAAW — Hallucination
  29. 29 HAAW — Grounding and Verification
  30. 30 HAAW — Tools and Agents
  31. 31 HAAW — The Agent Loop in Slow Motion
  32. 32 HAAW — AI Memory
  33. 33 HAAW — RWCX: Permissions and Control
  34. 34 HAAW — Multi-Agent Systems
  35. 35 HAAW — Coordination and Communication
  36. 36 HAAW — Failure, Recovery and Retry
  37. 37 HAAW — Planning and Replanning
  38. 38 HAAW — Goals, Constraints and Mission Success
  39. 39 HAAW — Authority, Escalation and Human Control
  40. 40 Autonomy — What Does It Actually Mean for an AI System to Act “On Its Own”?
  41. 41 Persistence and Long-Running Agents — What Happens When the Mission Lasts Hours, Days or Weeks?
  42. 42 Safety Boundaries and Sandboxes — How Do We Limit What an AI System Can Affect?
  43. 43 Monitoring, Audit and Observability — How Do We Know What an AI System Actually Did?
  44. 44 Reconstructing an AI Mission — From Event Records to a Defensible Account of What Happened
  45. 45 Reconstructing an AI Mission — From Event Records to a Defensible Account of What Happened