Unverified50% confidenceBenchmarkExact time
After scaling the data to 4K hours, JEPA rises to 49.3% on the tactile-to-video top-5 recall task, MAE remains below 9%, and VAE degrades to 3.1%.
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the change test after removing the records, Jev's TVD was 49%, Gemini's 43%, luna's 30%, and Haiku's 27%; the noise floors for repeated questions were 3%, 12%, 9%, and 9% respectively.63% similarUnverifiedThe cache hit rate for exact SHA256 hash caching of prompts is often less than 4%, because an extra space or punctuation mark will invalidate the cache.60% similarUnverifiedAfter replaying the same conversation three times, SME's performance rose from 92.5% to 95%, while the comparison solution LORAC dropped from a high position to 87.5%.59% similarUnverifiedIn task A, after reweighting by production volume, the 9B's allow proportion dropped from the 27B's 86.4% to 49.5%, clarify rose from 13.4% to 50.1%, and the weighted agreement rate was 61.2%.59% similarUnverified4 of the LoRAs scored lower than the no-LoRA control group, actually causing less motion in the image.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1152659API
curl https://kongchang.com/api/v1/knowledge/claims/1152659MCP
get_claim(id=1152659)