Unverified50% confidenceFactExact time
The paper "Weaken to Strengthen: Weakening-Induced On-Policy Self-Distillation for Large Reasoning Models" proposes the WISD method, which actively weakens the model and uses the contrast before and after weakening as a training signal, without requiring an external teacher or privileged information.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA paper titled "Self-Modeling Interventions Modulate Emergent Misalignment" proposes modulating the emergent misalignment phenomenon through self-modeling interventions.68% similarUnverifiedThe choice of weakening strength in WISD lacks theoretical grounding and may require tuning for different models and tasks.67% similarUnverifiedThe improvement ceiling of runtime improvement is constrained by the base model's inherent capabilities, and the model cannot break through the knowledge and reasoning boundaries obtained during pre-training through runtime optimization.67% similarVerifiedThe teacher model outputs in distillation training rarely include self-correction examples, causing student models to form an implicit belief that "strong models don't make mistakes," leading them to repeat rather than correct when verification fails.67% similarUnverifiedOver-alignment is essentially a product of bias in the alignment objective function, where the model has learned to avoid criticism rather than to be genuinely helpful.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114230API
curl https://kongchang.com/api/v1/knowledge/claims/1114230MCP
get_claim(id=1114230)