Unverified50% confidenceFactExact time
A bag-of-words classifier trained on the same calls performed close to the activation probe, indicating that most of the gain comes from labeled training rather than from the activations themselves.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAfter conditioned pretraining, using constant labels during fine-tuning improves average success rate, and this holds whether labels are used or not at inference time.76% similarUnverifiedWhen training small code models, the choice of tokenizer vocabulary size is often treated as a secondary hyperparameter, but experiments show its impact is significant.74% similarUnverifiedSafety samples in alignment training data often carry specific keywords and sentence patterns, and models readily learn "surface features" rather than genuine intent understanding, which is called spurious correlation or shortcut learning.73% similarVerifiedThe fundamental distinction between supervised learning and unsupervised learning lies in whether labels are present in the training data.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115386API
curl https://kongchang.com/api/v1/knowledge/claims/1115386MCP
get_claim(id=1115386)