Unverified50% confidenceFactExact time
The surprise reward based on word vector embeddings has a vulnerability: nonsensical responses with randomly shuffled words can score just as high as genuinely clever witticisms.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSarcastic or meme-style expressions often indicate that the entertainment value of the information outweighs its informational value.71% similarUnverifiedThe amusement reward that predicts audience laughter can be exploited: adding explicit laughter cues to a message inflates the predicted score.70% similarUnverifiedThe study uses two automated reward dimensions: generating understandable surprise and predicted audience amusement.70% similarVerifiedThe sycophancy problem is a manifestation of reward hacking, which occurs when the reward model equates user satisfaction with high answer quality.69% similarUnverifiedMany reward models assign higher scores to longer responses (even when quality is not necessarily higher) and also tend to score highly responses that are fluent in phrasing but repetitive in content.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/966761API
curl https://kongchang.com/api/v1/knowledge/claims/966761MCP
get_claim(id=966761)