How the MtG Color Wheel Explains AI Safety

Scott Garrabrant

15 Feb 2019 23:42 UTC

51 points

4 comments6 min readLW link

Alignment Newsletter #45

14 Feb 2019 2:10 UTC

25 points

2 comments LW link

Humans interpreting humans

Stuart_Armstrong

13 Feb 2019 19:03 UTC

10 points

1 comment LW link

Anchoring vs Taste: a model

Stuart_Armstrong

13 Feb 2019 19:03 UTC

11 points

0 comments LW link

Nuances with ascription universality

12 Feb 2019 23:38 UTC

16 points

1 comment LW link

Learning preferences by looking at the world

12 Feb 2019 22:25 UTC

45 points

10 comments7 min readLW link

(bair.berkeley.edu)

Would I think for ten thousand years?

Stuart_Armstrong

11 Feb 2019 19:37 UTC

25 points

12 comments LW link

“Normative assumptions” need not be complex

Stuart_Armstrong

11 Feb 2019 19:03 UTC

11 points

0 comments LW link

Coherent behaviour in the real world is an incoherent concept

11 Feb 2019 17:00 UTC

29 points

8 comments8 min readLW link

Some Thoughts on Metaphilosophy

10 Feb 2019 0:28 UTC

53 points

20 comments LW link

The Argument from Philosophical Difficulty

10 Feb 2019 0:28 UTC

47 points

18 comments LW link

Reinforcement Learning in the Iterated Amplification Framework

9 Feb 2019 0:56 UTC

24 points

7 comments LW link

HCH is not just Mechanical Turk

9 Feb 2019 0:46 UTC

36 points

4 comments LW link

Test Cases for Impact Regularisation Methods

6 Feb 2019 21:50 UTC

51 points

3 comments LW link

Security amplification

paulfchristiano

6 Feb 2019 17:28 UTC

20 points

0 comments LW link

Alignment Newsletter #44

6 Feb 2019 8:30 UTC

20 points

0 comments LW link

When to use quantilization

5 Feb 2019 17:17 UTC

52 points

5 comments LW link

Conclusion to the sequence on value learning

3 Feb 2019 21:05 UTC

44 points

13 comments LW link

[Question] How does Gradient Descent Interact with Goodhart?

Scott Garrabrant

2 Feb 2019 0:14 UTC

65 points

14 comments LW link

Reliability amplification

paulfchristiano

31 Jan 2019 21:12 UTC

21 points

3 comments LW link