Unverified50% confidenceFactExact time
Large models often exhibit overconfidence on open-domain questions, especially in domains where training data coverage is sparse.
1
Sources
50%
Confidence
Long-term
Relevance
11/1/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedData leakage falls into two categories—target leakage and train-test contamination—and leads to systematic overestimation of model performance.75% similarUnverifiedThe attack surface of frontier training environments grows nonlinearly with cluster scale, and low-cost penetration testing struggles to cover configuration errors and broken chains of trust at the scale of tens of thousands of GPUs.74% similarUnverifiedThe overconfidence of a model answering directly while skipping retrieval is technically referred to as miscalibration, meaning there is a systematic deviation between confidence and actual accuracy.74% similarUnverifiedGeneral-purpose large models often suffer from overconfidence, and using raw softmax probabilities directly as confidence is unreliable.74% similarVerifiedSome models carry the risk of over-optimization toward test sets, causing leaderboard scores to deviate from real-world performance (benchmark saturation).74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1123110API
curl https://kongchang.com/api/v1/knowledge/claims/1123110MCP
get_claim(id=1123110)