Unverified50% confidenceOpinionExact time
When a model faces out-of-distribution questions, its internal probability distribution is often too flat or too sharp, causing confidence to be systematically too high.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGeneral-purpose large models often suffer from overconfidence, and using raw softmax probabilities directly as confidence is unreliable.75% similarUnverifiedThe probability distribution learned by a model is naturally biased toward high-frequency events and common patterns, resulting in a systematic underestimation of long-tail events.74% similarUnverifiedTraditional privacy-preserving alignment methods usually assume that the reward model parameters are dense, causing the upper bound on estimation error to grow polynomially with the model dimension d.73% similarUnverifiedEquilibrium mismatch refers to the learned steady-state distribution itself being far from the target distribution; under certain explicit assumptions, even when the per-atom error in a locally symmetry-preserving model is very small, the distribution discrepancy still approaches its maximum value as the system size increases.73% similarUnverifiedWhen the guidance signal deviates from the model's own sampling distribution, the policy gradient estimate incurs bias.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084177API
curl https://kongchang.com/api/v1/knowledge/claims/1084177MCP
get_claim(id=1084177)