Unverified50% confidenceFactExact time
Measured on 45 real candidates, the similarity between the HRR query vector and the stored fact vectors fluctuated between -0.023 and +0.015, with a span of 0.038 constituting noise; after rescaling, the term's contribution to each score was 0.489–0.508, i.e., a near-constant 0.147–0.152 in the final result.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAblation analysis shows that fine-tuning on the segment paradigm yields a +5.29-point improvement on the test set, while replacing the first 4,500 transcript words with 2,000 retrieved words yields only a +1.55-point improvement.69% similarUnverifiedIn the fixed-vector self-similarity experiment, when the offset goes from 0 to 1024, the mean of the raw scores always fluctuates within ±0.38, while the mean of the absolute values only drops from 3.17 to 3.02, a change of about 4%.69% similarUnverifiedIn HDFS anomaly detection, Decision-1's raw output probabilities correlate 0.43 with event count and only 0.14 with labels; the model overreacts to lexical cues and log volume, and misjudges missing events as healthy.68% similarUnverifiedIn the author's tests, the fluctuation across all repeated runs is no more than 0.5% on the 3090 and at worst 2.6% on the L40S, so the data is not noise.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1062252API
curl https://kongchang.com/api/v1/knowledge/claims/1062252MCP
get_claim(id=1062252)