Unverified50% confidenceFactExact time
Experiments found that FIDE's preferences changed sensor selection, favoring designs that make the minimal correction to the benchmark model.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/30/2026
First Seen
Valid until: 1/28/2027
Sources
Related Entities
Related Claims
UnverifiedResearchers have found that fine-tuning only a small number of model parameters is sufficient to remove safety alignment (de-alignment attacks), and that certain jailbreak prompts can transfer across different models.73% similarUnverified底层模型更换后，即便微小的输出分布变化也可能导致此前校准的安全护栏过度拦截或防护失效，需要重新标注和调试70% similarUnverifiedFIDE offers a new perspective on experimental design: in benchmark-preserving settings, one should not seek to recover the "true" law, but rather to make minimal corrections on top of the existing benchmark.70% similarUnverified通过'不要按常规套路'这样的约束性提示词，本质上是在调整模型的采样分布，降低高频方案的输出权重，迫使模型探索低概率但可能更具创新性的输出区域70% similarUnverifiedStandard Bayesian optimization acquisition functions (such as GP-UCB, EI, Thompson sampling) are almost all designed to minimize cumulative regret, and therefore naturally favor exploitation over exploration.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114457API
curl https://kongchang.com/api/v1/knowledge/claims/1114457MCP
get_claim(id=1114457)