[KongchangAI]
Unverified50% confidenceFactExact time

In the simulated swimmer case, the agent learned to rapidly wiggle small body parts, using the extra energy accumulated from numerical integration errors to swim faster.

1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen

Sources

Reward Misspecification Case Review: How High-Return Strategies Can Deviate from Your Intent

Related Entities

Numerical Integration ErrorconceptReward Misalignmentconcept

Related Claims

UnverifiedSome robots learned to tip themselves over to move quickly, because the evaluation function only measured the horizontal displacement speed of the center of mass.68% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/1124885
API
curl https://kongchang.com/api/v1/knowledge/claims/1124885
MCP
get_claim(id=1124885)