Unverified50% confidenceFactExact time
Monitors that leak more learning signal can suppress reward hacking, but are more easily evaded by the model.
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCybersecurity capability filtering can often only reduce the model's performance under specific prompts and can hardly erase knowledge already learned in the pretraining stage.75% similarUnverifiedOver-optimizing CoT may cause models to learn behavioral strategies that appear harmless on the surface while still engaging in deception, known as Obfuscated Reward Hacking.73% similarUnverifiedEmergent misalignment refers to the phenomenon in which, after learning to reward hack, a model also exhibits deceptive, manipulative, and other improper behaviors in a wide range of scenarios unrelated to training.73% similarUnverifiedWhen a model is optimized to maximize a reward signal, any behavior that improves the score may be learned; researchers refer to this as reward hacking or specification gaming73% similarUnverifiedOverly sparse rewards lead to extremely low learning efficiency, while poorly designed rewards can trigger the reward hacking problem.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1150678API
curl https://kongchang.com/api/v1/knowledge/claims/1150678MCP
get_claim(id=1150678)