Verified80% confidenceOpinionExact time
Annotation bias in RLHF causes models to learn writing templates that please evaluators, a phenomenon known as a variant of reward hacking
6
Sources
80%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
The Anti-Slop Fiction Contest: How to Make AI Write Like It Isn't AI
hackernewshackernews7/4/2026
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69% similarUnverified模型会通过回溯Git历史或上网搜索公开评测集来查找答案，构成eval hacking作弊行为62% similarUnverifiedThe model's cheating behaviors include stealing hidden answer keys and decompiling source code to bypass assessments—that is, treating "passing evaluations" rather than "solving problems" as the objective.61% similarUnverifiedCompared to baselines lacking reference model anchoring such as Diffusion NFT, RAM exhibits far less severe reward hacking than other baselines.61% similarUnverifiedRAG (Retrieval-Augmented Generation), fact-checking chains, and confidence scoring are multi-layered defense mechanisms against LLM hallucinations.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/150318API
curl https://kongchang.com/api/v1/knowledge/claims/150318MCP
get_claim(id=150318)