AI ALIGNMENT FORUM
AF

peligrietzer

Posts

Sorted by New

0peligrietzer's Shortform

2y

0

21The Problem With the Word ‘Alignment’

6mo

2

37Paper: Understanding and Controlling a Maze-Solving Policy Network

1y

0

22Behavioural statistics for a maze-solving agent

2y

10

37Maze-solving agents: Add a top-right vector, make the agent go to the top-right

2y

7

139Understanding and controlling a maze-solving policy network

2y

22

47Predictions for shard theory mechanistic interpretability results

2y

6

11[Simulators seminar sequence] #2 Semiotic physics - revamped

2y

14

19 [Simulators seminar sequence] #1 Background & shared assumptions

2y

3

27A Short Dialogue on the Meaning of Reward Functions

2y

0

Wiki Contributions

Comments

Sorted by

Behavioural statistics for a maze-solving agent

peligrietzer2y10

I'd maybe point the finger more at the simplicity of the training task than at the size of the network? I'm not sure there's strong reason to believe the network is underparameterized for the training task. But I agree that drawing lessons from small-ish networks trained on simple tasks requires caution.