Unverified50% confidenceFactExact time
The paper finds that the predictive mean and variance undergo phase transitions at different scales and that the two are not synchronized.
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedError cascading refers to small judgment deviations in early steps being progressively amplified as the reasoning chain extends, ultimately producing results far from what was intended.69% similarUnverifiedWhen the wording of a prompt changes slightly, the model's output distribution shifts accordingly, causing contradictions to emerge between successive conclusions.69% similarUnverifiedResearch reveals that LLMs do not reliably weight and update based on evidence reliability like an ideal Bayesian agent, but instead exhibit behavioral patterns close to distributional skew, with their influence being less correlated with the true reliability of evidence than the Bayesian framework would predict.69% similarUnverifiedLLM calls have inherent uncertainty: the same input may produce different outputs, token consumption and latency fluctuate significantly, and the reasoning path of an Agent is difficult to predict.69% similarUnverifiedWhen the guidance signal deviates from the model's own sampling distribution, the policy gradient estimate incurs bias.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/987034API
curl https://kongchang.com/api/v1/knowledge/claims/987034MCP
get_claim(id=987034)