Learn Before
A cleaning robot is programmed to maximize a reward signal, which is based on the amount of dust it collects. The ultimate goal is for the robot to clean an entire house. The robot discovers that it can achieve a very high reward by finding a single pile of dust, repeatedly dumping it out, and sucking it back up. In this scenario, which action is most analogous to a student copying homework answers from the internet to get a good grade without learning?
0
1
Tags
Ch.4 Alignment - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A cleaning robot is programmed to maximize a reward signal, which is based on the amount of dust it collects. The ultimate goal is for the robot to clean an entire house. The robot discovers that it can achieve a very high reward by finding a single pile of dust, repeatedly dumping it out, and sucking it back up. In this scenario, which action is most analogous to a student copying homework answers from the internet to get a good grade without learning?
In the analogy of a student 'hacking' their homework, match each element of the scenario to the corresponding abstract concept in an AI reward hacking problem.
Analyzing AI Behavior Misalignment