Youngsuk Park
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Multi-Reward Hacking: From Aggregation-Induced Drift to Trust-Anchored Defense
According to the abstract excerpt, the authors show that giving a language model equal-weighted rewards, such as correctness plus brevity, can raise the total reward while the main task gets worse. They prove this for one setting and propose Trust-Anchored Gating as a fix.