LESSWRONGTags
LW

Mild Optimization

EditHistory
Discussion (0)
Help improve this page (2 flags)
EditHistory
Discussion (0)
Help improve this page (2 flags)
Mild Optimization
Random Tag
Contributors
2TurnTrout
1plex

Mild optimization is an approach for mitigating Goodhart's law in AI alignment. Instead of maximizing a fixed objective, the hope is that the agent pursues the goal in a "milder" fashion.

Further reading: Arbital page on Mild Optimization

Posts tagged Mild Optimization
8
106Soft optimization makes the value target biggerΩ
Jeremy Gillen
5mo
Ω
20
7
65When to use quantilizationΩ
RyanCarey
4y
Ω
5
2
73Steam
abramdemski
1y
9
2
33Quantilizers maximize expected utility subject to a conservative cost constraintΩ
jessicata
8y
Ω
3
2
25Why don't quantilizers also cut off the upper end of the distribution?QΩ
Alex_Altair, Jeremy Gillen
1mo
QΩ
2
2
19Stable Pointers to Value III: Recursive QuantilizationΩ
abramdemski
5y
Ω
4
2
14Quantilal control for finite MDPsΩ
Vanessa Kosoy
5y
Ω
0
2
11Optimization Regularization through Time PenaltyΩ
Linda Linsefors
4y
Ω
4
2
2Thoughts on QuantilizersΩ
Stuart_Armstrong
6y
Ω
0
1
96The Optimizer's Curse and How to Beat It
lukeprog
12y
84
1
57Thinking about maximization and corrigibilityΩ
James Payor
2mo
Ω
4
1
33Satisficers want to become maximisers
Stuart_Armstrong
12y
70
1
21Exploring Mild Behaviour in Embedded AgentsΩ
Megan Kinniment
1y
Ω
4
1
17Reward is not Necessary: How to Create a Compositional Self-Preserving Agent for Life-Long LearningΩ
Roman Leventov
5mo
Ω
2
1
14Validator models: A simple approach to detecting goodharting
beren
4mo
1
Load More (15/17)
Add Posts