Unverified50% confidenceFactExact time
Under severe pruning, the model can generate diverse chain-of-thought trajectories, but this is accompanied by increased repetition and reduced generation reliability.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedDuring model collapse, the model successively experiences a reduction in output diversity, loss of minority categories and edge knowledge, and ultimately produces meaningless or highly repetitive outputs.80% similarUnverifiedAs model capabilities improve, the tension between capability and alignment becomes increasingly pronounced, and the consequences of alignment failure are correspondingly amplified.74% similarVerifiedWhen a single large model independently completes long-chain tasks requiring more than dozens of reasoning steps, errors accumulate exponentially with chain length, whereas decomposing the task among multiple specialized agents can significantly reduce the error rate.74% similarUnverifiedThe vaguer the prompt, the more the model output collapses toward high-probability regions, and the stronger the consistency.74% similarUnverifiedThe higher the heterogeneity, the harder it is for the global model obtained through federated aggregation to behave consistently across clients, and the more inconsistent the gradient directions, which is a potential mechanism by which explanation maps may become distorted.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129107API
curl https://kongchang.com/api/v1/knowledge/claims/1129107MCP
get_claim(id=1129107)