以下部分引用 AI 总结，现阶段 AI 仍然有幻觉。请以内容的原文为准。
数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15
WBench
9 atoms · 跨 3 天 · 首见 2026-05-26 · 最近 2026-06-05
三色: 🟦 fact 7 · 🟥 take 2 · stance ▲0/▼0/◆4
来源: X 9
时态: fresh:9
标签: 好数字:5 · 好观点:3 · 好思考:2 · 好信源:1
叙事 (narrative) (4)
- 2026-05-27 · 规模
289 core test cases and 1,058 interaction rounds across diverse scenes, styles, and subjects.
tags: 好数字 · → daily
value: qty=289个核心测试用例和1,058轮互动 · date=2026-05-27
- 2026-05-27 · 自动评分与人类评判的相关性
Automatic scoring achieves a Spearman correlation of at least 0.94 with human blind pairwise win rates, with some metrics reaching a perfect 1.0.
tags: 好数字 · → daily
value: qty=斯皮尔曼相关系数≥0.94，部分指标达1.0 · date=2026-05-27
- 2026-06-05 · 范围
WBench, a high-quality world generation benchmark for Interactive Video World Model Evaluation
tags: 好观点 · → daily
- 2026-06-05 · 评估维度
Benchmarking: 1. video quality 2. Consistency 3. Physics compliance 4. Whether the model generates what it's told (important)
tags: 好观点 · → daily
事实 (fact) (5)
- 2026-05-26 · 22 metrics across 5 dimensions reveal no single model dominates
22 metrics across 5 dimensions reveal a key finding. No single model dominates quality, control, and physics.
tags: 好数字·好思考 · → daily
value: qty=22 metrics, 5 dimensions
- 2026-05-26 · every model has blind spots
Every model has blind spots.
tags: 好思考·好观点 · → daily
- 2026-05-26 · benchmarks 20 video world models
WBench benchmarks 20 video world models
tags: 好数字 · → daily
value: qty=20
- 2026-05-26 · multi-turn evaluation with 289 cases and 1,058 turns
A comprehensive multi-turn evaluation with 289 cases and 1,058 turns.
tags: 好数字 · → daily
value: qty=289 cases, 1058 turns
- 2026-05-26 · tested Kling 3.0, Wan 2.7, Seedance 1.5, Genie 3, and 16 others
We tested Kling 3.0, Wan 2.7, Seedance 1.5, Genie 3, and 16 others.
tags: 好信源 · → daily
value: qty=20 models
时间轴 (近 20)
- 2026-06-05 ·
narrative · 范围 · → daily
- 2026-06-05 ·
narrative · 评估维度 · → daily
- 2026-05-27 ·
narrative · 规模 · → daily
- 2026-05-27 ·
narrative · 自动评分与人类评判的相关性 · → daily
- 2026-05-26 ·
fact · benchmarks 20 video world models · → daily
- 2026-05-26 ·
fact · multi-turn evaluation with 289 cases and 1,058 turns · → daily
- 2026-05-26 ·
fact · 22 metrics across 5 dimensions reveal no single model domina · → daily
- 2026-05-26 ·
fact · tested Kling 3.0, Wan 2.7, Seedance 1.5, Genie 3, and 16 oth · → daily
- 2026-05-26 ·
fact · every model has blind spots · → daily
← 实体目录 · 系统日志