Unverified50% confidenceFactExact time
Adjusting the subgroup proportions in the training data may work, whereas simply increasing the amount of data may not solve the subgroup shift problem.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedSimply stacking parameters without sufficient training data results in a model that is large but not capable; a leap in parameters must be accompanied by a proportionally expanded training corpus and computing power to realize capability gains.79% similarVerifiedAlignment training optimizes the model's statistical behavior in known scenarios rather than establishing formally verifiable security boundaries, and may fail when faced with rare instruction combinations or role-playing frameworks outside the training distribution.79% similarUnverifiedMinority-class-prioritized data collection brings some benefit when the training set is very small, but its advantage does not persist as the amount of data increases.78% similarUnverifiedThe generalization discussed in this experiment is closer to out-of-distribution reasoning, i.e., the training data distribution does not contain weight-function paired text.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1154578API
curl https://kongchang.com/api/v1/knowledge/claims/1154578MCP
get_claim(id=1154578)