Charlie Steiner(Charlie Steiner)

Karma: 5,085

LW1.0 username Manfred. PhD in condensed matter physics. I am independently thinking and writing about value learning.

[Simulators seminar sequence] #2 Semiotic physics—revamped

Jan, Charlie Steiner, Logan Riggs, janus, jacquesthibs, metasemi, Michael Oesterle, Lucas Teixeira, peligrietzer and remember

27 Feb 2023 0:25 UTC

20 points

22 comments13 min readLW link

Shard theory alignment has important, often-overlooked free parameters.

Charlie Steiner20 Jan 2023 9:30 UTC

32 points

10 comments3 min readLW link

[Simulators seminar sequence] #1 Background & shared assumptions

Jan, Charlie Steiner, Logan Riggs, janus, jacquesthibs, metasemi, Michael Oesterle, Lucas Teixeira, peligrietzer and remember

2 Jan 2023 23:48 UTC

46 points

4 comments3 min readLW link

Take 14: Corrigibility isn’t that great.

Charlie Steiner25 Dec 2022 13:04 UTC

15 points

3 comments3 min readLW link

Take 13: RLHF bad, conditioning good.

Charlie Steiner22 Dec 2022 10:44 UTC

53 points

4 comments2 min readLW link

Take 12: RLHF’s use is evidence that orgs will jam RL at real-world problems.

Charlie Steiner20 Dec 2022 5:01 UTC

25 points

1 comment3 min readLW link

Take 11: “Aligning language models” should be weirder.

Charlie Steiner18 Dec 2022 14:14 UTC

31 points

0 comments2 min readLW link

Take 10: Fine-tuning with RLHF is aesthetically unsatisfying.

Charlie Steiner13 Dec 2022 7:04 UTC

36 points

3 comments2 min readLW link

Take 9: No, RLHF/IDA/debate doesn’t solve outer alignment.

Charlie Steiner12 Dec 2022 11:51 UTC

33 points

14 comments2 min readLW link

Take 8: Queer the inner/outer alignment dichotomy.

Charlie Steiner9 Dec 2022 17:46 UTC

28 points

3 comments2 min readLW link

Take 7: You should talk about “the human’s utility function” less.

Charlie Steiner8 Dec 2022 8:14 UTC

47 points

22 comments2 min readLW link

Take 6: CAIS is actually Orwellian.

Charlie Steiner7 Dec 2022 13:50 UTC

14 points

5 comments2 min readLW link

Take 5: Another problem for natural abstractions is laziness.

Charlie Steiner6 Dec 2022 7:00 UTC

30 points

4 comments3 min readLW link

Take 4: One problem with natural abstractions is there’s too many of them.

Charlie Steiner5 Dec 2022 10:39 UTC

36 points

4 comments1 min readLW link

Take 3: No indescribable heavenworlds.

Charlie Steiner4 Dec 2022 2:48 UTC

22 points

12 comments2 min readLW link

Take 2: Building tools to help build FAI is a legitimate strategy, but it’s dual-use.

Charlie Steiner3 Dec 2022 0:54 UTC

17 points

1 comment2 min readLW link

Take 1: We’re not going to reverse-engineer the AI.

Charlie Steiner1 Dec 2022 22:41 UTC

38 points

4 comments4 min readLW link

Some ideas for epistles to the AI ethicists

Charlie Steiner14 Sep 2022 9:07 UTC

19 points

0 comments4 min readLW link

The Solomonoff prior is malign. It’s not a big deal.

Charlie Steiner25 Aug 2022 8:25 UTC

38 points

9 comments7 min readLW link

Reducing Goodhart: Announcement, Executive Summary

Charlie Steiner20 Aug 2022 9:49 UTC

14 points

0 comments1 min readLW link