LESSWRONG
LW

2022 MIRI Alignment Discussion

Jun 15, 2022 by Rob Bensinger

A collection of MIRI write-ups and conversations about alignment released in 2022, following the Late 2021 MIRI Conversations.

317Six Dimensions of Operational Adequacy in AGI Projects
Ω
Eliezer Yudkowsky
3y
Ω
66
946AGI Ruin: A List of Lethalities
Ω
Eliezer Yudkowsky
3y
Ω
711
273A central AI alignment problem: capabilities generalization, and the sharp left turn
Ω
So8res
3y
Ω
55
314On how various plans miss the hard bits of the alignment challenge
Ω
So8res
3y
Ω
89
173The inordinately slow spread of good AGI conversations in ML
Rob Bensinger
3y
62
197A note about differential technological development
Ω
So8res
3y
Ω
33
136Brainstorm of things that could force an AI team to burn their lead
Ω
So8res
3y
Ω
8
177AGI ruin scenarios are likely (and disjunctive)
Ω
So8res
3y
Ω
38
65Where I currently disagree with Ryan Greenblatt’s version of the ELK approach
Ω
So8res
3y
Ω
7
159Why all the fuss about recursive self-improvement?
So8res
3y
62
50Humans aren't fitness maximizers
Ω
So8res
3y
Ω
46
130Warning Shots Probably Wouldn't Change The Picture Much
Ω
So8res
3y
Ω
42
74What does it mean for an AGI to be 'safe'?
Ω
So8res
3y
Ω
29
136Don't leave your fingerprints on the future
Ω
So8res
3y
Ω
48
134Niceness is unnatural
Ω
So8res
3y
Ω
20
105Contra shard theory, in the context of the diamond maximizer problem
Ω
So8res
3y
Ω
19
64Notes on "Can you control the past"
Ω
So8res
3y
Ω
41
171Decision theory does not imply that we get to have nice things
Ω
So8res
3y
Ω
73
132Superintelligent AI is necessary for an amazing future, but far from sufficient
Ω
So8res
3y
Ω
48
111How could we know that an AGI system will have good consequences?
Ω
So8res
3y
Ω
25
72Distinguishing test from training
Ω
So8res
3y
Ω
11
302A challenge for AGI organizations, and a challenge for readers
Ω
Rob Bensinger, Eliezer Yudkowsky
3y
Ω
33
102Thoughts on AGI organizations and capabilities work
Ω
Rob Bensinger, So8res
3y
Ω
17