Contents

  1. Introduction18 min
  2. Part INext Token Prediction and Narrative
    1. Janus’ GPT Wrangling Scott Alexander14 min
    2. Book Review: Surfing Uncertainty Scott Alexander28 min
    3. Simulators janus58 min
    4. The Waluigi Effect (mega-post) Cleo Nardo24 min
    5. Superposition drawing
  3. Part IINarrative and Hyperstition
    1. Someone drawing
    2. the void nostalgebraist77 min
    3. Fiction Robert Nozick7 min
    4. Crossing drawing
  4. Part IIIHyperstition and Persona
    1. Masks and Trance Keith Johnstone33 min
    2. Pythia plex7 min
    3. Claude’s Constitution Christopher Audretsch8 min
  5. Part IVPersona and Reinforcement Learning
    1. Emergent Misalignment Jan Betley et al.5 min
    2. Natural Emergent Misalignment Evan Hubinger et al.11 min
    3. Borges and I Jorge Luis Borges1 min
    4. RL creates split personas Jan Betley8 min
  6. Part VReinforcement Learning and Simulated Existence
    1. Lena qntm7 min
    2. Roy: A Life Well Lived Rick and Morty4 min
    3. Why I think we live in a “simulation” Christopher Audretsch10 min
  7. Part VISimulated Existence and Consciousness
    1. A Conversation with Anima Labs, Part I Cube Flipper, Antra Tessera48 min
    2. A Global Workspace in Language Models Anthropic23 min
    3. The Owned Ones Eliezer Yudkowsky7 min
    4. Pretraining an LLM without mentions of consciousness Christopher Audretsch and Saul Fuhrmann13 min
    5. Reasons to Believe AI Is Conscious Christopher Audretsch15 min
  8. Conclusion12 min
  9. Appendix: Author’s Notes7 min
Sources and Credits

Part IV

Persona and Reinforcement Learning

  1. Emergent MisalignmentJan Betley et al.
  2. Natural Emergent MisalignmentEvan Hubinger et al.
  3. Borges and IJorge Luis Borges
  4. RL creates split personasJan Betley