Unverified50% confidenceFactExact time
For molecules whose correct values the models could recall, the LLMs reported lower uncertainty, indicating that the models exhibit overconfidence when relying on memory.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLLM输出的置信度来自logits或token概率，是模型内部的相对偏好，并不等同于事件发生的真实频率，即模型可能过度自信。73% similarVerifiedThe root cause of LLM hallucination lies in the autoregressive generation mechanism—the model predicts the next Token based on a probability distribution rather than genuine logical reasoning.73% similarUnverifiedChanging the reasoning effort (e.g., requiring step-by-step reasoning or adjusting the computation budget) had little effect on LLM accuracy on the molecular property prediction task.72% similarUnverifiedA memory state that accurately reproduces all predictions of the current model may have already lost the ability to continue learning from the next observation.70% similarVerifiedCalibration means that the probabilities output by the model should match reality; neural networks generally have an overconfidence tendency, so model probabilities should not be blindly trusted, and thresholds should be set only after validation with labeled data.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1124963API
curl https://kongchang.com/api/v1/knowledge/claims/1124963MCP
get_claim(id=1124963)