Encyclopedia

Notes

6 published Notes

Reading guides and evidence-led synthesis. Each Note connects a question to the research and historical record.

• note

  AI Incidents: What Failures Actually Establish

  Real failures reveal where safeguards break; comparing investigations, participant accounts and evaluations shows which explanations the evidence supports.

• note

  From objectives to accountable control

  Why specifying goals, checking behavior, intervening and preserving human responsibility require different evidence.

• note

  Key Papers: A Reading Guide

  Question-led reading sequences through alignment research, with reasons for each step and prompts for assessing the evidence.

• note

  Origins of Alignment: Problems Before the Name

  Older work separates intelligent performance, communicating purposes and reliable delegation; later research makes some of these problems explicit and testable.

• note

  When a safety evaluation becomes a forecast

  How sampling, measurement and post-release checks change what a safety test can say about future use.

• note

  Who Supplies Alignment Feedback?

  When does more alignment feedback improve supervision, and when does it leave error detection or representation unresolved?