Unverified50% confidenceBenchmarkExact time
After cleaning up the anchors and prompts, promptfoo's agreement rate rose from 34/40 to 37/40, and DeepEval's rose from 35/40 to 36/40.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedIn tests on 2,456 held-out prompts, increasing the sampling steps from 20 to 50 improved FID only from 27.0 to 26.6 (a gain of 0.4), while removing the shift caused FID to rise from 27.0 to 27.3 and FD-DINOv2 to rise from 218 to 228.68% similarUnverifiedThe first run of the anchor rewrite produced split results: all four segments of the subtitle footage improved by 3–4dB (37.3/32.3/28.7/30.1dB), while the masked-region PSNR of the gradual-drift footage dropped from 38.30 to 34.69dB, falling below the stop line.65% similarUnverifiedUnder an 18k character budget, anchored hybrid retrieval achieved 36.18% accuracy, BM25 achieved 35.98%, while the best direct planner-guided method reached only 34.19%.61% similarUnverifiedExperience internalization fine-tuning improved Qwen3-8B-27B's one-shot score from 37.07 to 48.22, while reducing generated tokens from 6.41 million to 570,000 (an 11-fold reduction).60% similarUnverifiedAfter fine-tuning, Laya's accuracy on the MASSIVE dataset can be improved from 0.45 to 0.85, and on another task from 0.362 to 0.766.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1018299API
curl https://kongchang.com/api/v1/knowledge/claims/1018299MCP
get_claim(id=1018299)