AI ALIGNMENT FORUMShard Theory
AF

Shard Theory

Written by Quintin Pope, Alex Turner, Charles Foster, and Logan Smith. Card image generated by DALL-E 2:

53Humans provide an untapped wealth of evidence about alignment

Alex Turner, Quintin Pope

8mo

28

41Human values & biases are inaccessible to the genome

9mo

25

25General alignment properties

8mo

1

26Evolution is a bad analogy for AGI: inner alignment

8mo

0

84Reward is not the optimization target

8mo

77

69The shard theory of human values

Quintin Pope, Alex Turner

6mo

30

21Understanding and avoiding value drift

7mo

5

33A shot at the diamond-alignment problem

6mo

35

31Don't design agents which exploit adversarial inputs

Alex Turner, Garrett Baker

5mo

27

24Don't align agents to evaluations of plans

4mo

28

28Alignment allows "nonrobust" decision-influences and doesn't require robust grading

4mo

31

44Inner and outer alignment decompose one hard problem into two extremely hard problems

4mo

11