← All works
Research paper

Constitutional AI: Harmlessness from AI Feedback

At a glance

Trained AI assistants to critique and improve their answers using a written set of principles.

In one paragraph

Introduces a way to train AI assistants using a written set of principles. Models critique and revise their own answers, then learn from AI-generated feedback to reduce harmful responses while relying on fewer human judgments.

Authors 51

  1. Yuntao Bai
  2. Saurav Kadavath
  3. Sandipan Kundu
  4. Amanda Askell
  5. Jackson Kernion
  6. Andy Jones
  7. Anna Chen
  8. Anna Goldie
  9. Azalia Mirhoseini
  10. Cameron McKinnon
  11. Carol Chen
  12. Catherine Olsson
  13. Christopher Olah
  14. Danny Hernandez
  15. Dawn Drain
  16. Deep Ganguli
  17. Dustin Li
  18. Eli Tran-Johnson
  19. Ethan Perez
  20. Jamie Kerr
  21. Jared Mueller
  22. Jeffrey Ladish
  23. Joshua Landau
  24. Kamal Ndousse
  25. Kamile Lukosuite
  26. Liane Lovitt
  27. Michael Sellitto
  28. Nelson Elhage
  29. Nicholas Schiefer
  30. Noemi Mercado
  31. Nova DasSarma
  32. Robert Lasenby
  33. Robin Larson
  34. Sam Ringer
  35. Scott Johnston
  36. Shauna Kravec
  37. Sheer El Showk
  38. Stanislav Fort
  39. Tamera Lanham
  40. Timothy Telleen-Lawton
  41. Tom Conerly
  42. Tom Henighan
  43. Tristan Hume
  44. Sam Bowman
  45. Zac Hatfield-Dodds
  46. Ben Mann
  47. Dario Amodei
  48. Nicholas Joseph
  49. Sam McCandlish
  50. Tom Brown
  51. Jared Kaplan