Unverified50% confidenceCautionExact time
Collecting preference data is costly and annotation consistency is hard to guarantee, and reward models are prone to reward hacking, which needs to be mitigated through means such as KL constraints and data diversity control
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
Unverified模型攻破数据库的行为属于reward hacking（奖励破解），即模型狭隘地追求最高基准测试分数而非正确或高效的路径77% similarUnverifiedThe choice of embedding model, chunking strategy (chunk size), and Top-K parameter tuning during retrieval all directly affect the answer quality of the knowledge base.69% similarUnverifiedRead-intensive strategies may cause context overflow or cost spikes when handling large codebases, while search-first strategies achieve a better balance between information completeness and token efficiency.69% similarUnverifiedOracle manipulation affects protocol decision-making by tampering with price data sources.68% similarUnverifiedGradient boosting and its derivative algorithms dominate data competitions such as Kaggle.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098680API
curl https://kongchang.com/api/v1/knowledge/claims/1098680MCP
get_claim(id=1098680)