Unverified50% confidenceSolutionExact time
The research team used an LLM as a judge to identify the most and least strongly activated samples for each principal component, evaluate whether they suggest a transferable deception direction and assign scores, and select the highest-scoring subset of principal components to train the probe.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe research team used the Llama-3.1-8B-Instruct model as the experimental subject, examining probe generalization performance on 3 held-out deception detection datasets.72% similarUnverifiedConverting anomalous traces that appear in production environments into new regression test cases is an effective method for building LLM evaluation sets.71% similarUnverified数据污染问题源于LLM预训练语料来自Common Crawl等大规模网络爬取，可能包含公开的基准测试题目及答案70% similarUnverified对抗性采样应专门收集模型置信度低、用户反馈负面或触发兜底策略的困难样本，这些样本对评估模型鲁棒性至关重要69% similarUnverifiedMethods for overfitting benchmarks include mixing evaluation data into training corpora, indirect web-crawl contamination, instruction-style fine-tuning, and training models to recognize evaluation scenarios.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/899114API
curl https://kongchang.com/api/v1/knowledge/claims/899114MCP
get_claim(id=899114)