Unverified50% confidenceFactExact time
The paper uses predictive representations to measure the similarity among values in multi-value alignment objectives, and finds that this similarity is significantly correlated with model robustness.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe paper "Predicting Alignment Generalization with Value Representations" proposes the alignment generalization prediction task, that is, predicting the behavioral changes of a fine-tuned model on a large number of held-out values after it follows a certain value.72% similarUnverifiedThe paper argues that joint modeling theoretically outperforms methods that coarsen data beforehand on preference comparison tasks, due to retaining richer information.71% similarUnverifiedSimile emphasizes modeling through analogy or similarity, inferring the reactions of similar populations based on the behavioral characteristics of known populations70% similarUnverifiedThe paper proves that pairwise comparison models and listwise comparison models are special cases of its joint model under specific constraints.70% similarUnverifiedThe study derives a first-order formula for predicting the expected divergence between predictors fitted on two independent corpora, accounting for variable document length, random context counts, and correlations from cross-token offsets.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013412API
curl https://kongchang.com/api/v1/knowledge/claims/1013412MCP
get_claim(id=1013412)