This website requires javascript to properly function. Consider activating javascript to get access to all site functionality.
AI ALIGNMENT FORUM
Alignment Hot Take Advent Calendar
AF
Login
Alignment Hot Take Advent Calendar
15
Take 1: We're not going to reverse-engineer the AI.
Charlie Steiner
1y
1
12
Take 2: Building tools to help build FAI is a legitimate strategy, but it's dual-use.
Charlie Steiner
1y
0
10
Take 3: No indescribable heavenworlds.
Charlie Steiner
1y
0
16
Take 4: One problem with natural abstractions is there's too many of them.
Charlie Steiner
1y
1
13
Take 5: Another problem for natural abstractions is laziness.
Charlie Steiner
1y
1
6
Take 6: CAIS is actually Orwellian.
Charlie Steiner
1y
2
24
Take 7: You should talk about "the human's utility function" less.
Charlie Steiner
1y
1
13
Take 8: Queer the inner/outer alignment dichotomy.
Charlie Steiner
1y
0
23
Take 9: No, RLHF/IDA/debate doesn't solve outer alignment.
Charlie Steiner
1y
5
16
Take 10: Fine-tuning with RLHF is aesthetically unsatisfying.
Charlie Steiner
1y
1
15
Take 11: "Aligning language models" should be weirder.
Charlie Steiner
1y
0
11
Take 12: RLHF's use is evidence that orgs will jam RL at real-world problems.
Charlie Steiner
1y
1
23
Take 13: RLHF bad, conditioning good.
Charlie Steiner
1y
0
10
Take 14: Corrigibility isn't that great.
Charlie Steiner
1y
3