How to Keep Improving When You're Better Than Any Teacher - Iterated Distillation and Amplification

[2nd upload] AI systems can be trained using demonstrations from experts, but how do you train them to out-perform those experts? Can this still be done even without clear win/loss criteria? And how do you do it safely?
This video was based on work including:
"Supervising strong learners by amplifying weak experts" by Paul Christiano, Buck Shlegeris, Dario Amodei (arxiv.org/abs/1810.08575)
openai.com/blog/amplifying-ai...
www.alignmentforum.org/s/EmDu...
ai-alignment.com/iterated-dis...
With thanks to my wonderful Patrons: ( / robertskmiles )
Steef
Jason Strack
Jordan Medina
Jason Hise
Scott Worley
JJ Hepboin
Pedro A Ortega
Said Polat
Chris Canal
Nicholas Kees Dupuis
James
Richárd Nagyfi
Phil Moyer
Alec Johnson
Clemens Arbesser
Bryce Daifuku
Simon Strandgaard
Jonatan R
Michael Greve
The Guru Of Vision
Volodymyr
David Tjäder
Julius Brash
Tom O'Connor
Erik de Bruijn
Robin Green
Laura Olds
Jon Halliday
Paul Hobbs
Jeroen De Dauw
Tim Neilson
Eric Scammell
Igor Keller
Ben Glanton
Robert Sokolowski
anul kumar sinha
Jérôme Frossard
Sean Gibat
Sun Sun
andrew Russell
Cooper Lawton
Gladamas
Sylvain Chevalier
DGJono
robertvanduursen
Dmitri Afanasjev
Brian Sandberg
Einar Ueland
Marcel Ward
Andrew Weir
Taylor Smith
Ben Archer
Scott McCarthy
Kabs Kabs Kabs
Tendayi Mawushe
Jannik Olbrich
Anne Kohlbrenner
Bjorn Nyblad
Jussi Männistö
Mr Fantastic
Wr4thon
Archy de Berker
Marc Pauly
Joshua Pratt
Shevis Johnson
Andy Kobre
Brian Gillespie
Martin Wind
Peggy Youell
Poker Chen
Kees
Darko Sperac
Truls
Paul Moffat
Jelle Langen
Anders Öhrt
Marco Tiraboschi
Michael Kuhinica
Fraser Cain
Robin Scharf
Oren Milman
John Rees
Shawn Hartsock
Seth Brothwell
Brian Goodrich
Clark Mitchell
Kasper Schnack
Michael Hunter
Klemen Slavic
Patrick Henderson
Long Nguyen
Oct todo22
Melisa Kostrzewski
Hendrik
Daniel Munter
Graham Henry
Duncan Orr