WEBVTT

1
00:00:00.280 --> 00:00:07.820
Which moved first, the blue circle or the yellow square? The
question alone does not specify either event order.
蓝圆还是黄方块先移动？仅凭问题不能确定事件顺序。

2
00:00:08.760 --> 00:00:14.820
With both frames hidden, an answer could be a guess; a correct
guess would not establish that anything was observed.
两帧隐藏时，答案可能是猜测；猜对也不能证明观察过视频。

3
00:00:15.980 --> 00:00:22.400
Now reveal two original event views: the blue circle moves in
the first, before the yellow square changes position.
现在揭示原创事件视图：先是蓝圆移动，随后黄方块改变位置。

4
00:00:23.420 --> 00:00:29.460
The visible sequence supports blue-circle first, linking the
answer to evidence rather than object expectations.
可见顺序支持蓝圆答案，让判断依赖具体证据，而不是对物体的常见预期。

5
00:00:30.680 --> 00:00:36.480
These are neutral geometric events, not footage of real people
or a recreated question from a research benchmark.
这是中性几何事件，不是真人视频，也不是重建研究基准题。

6
00:00:37.500 --> 00:00:42.480
When a model answers correctly, did the video supply information
necessary to reach that answer?
这个练习引出核心问题：模型答对时，视频是否真正提供了必要信息？

7
00:00:43.100 --> 00:00:49.320
Compare that event-order question with asking what color a shape
becomes after it is painted with red paint.
把事件顺序问题与“形状涂上红漆后变成什么颜色”比较。

8
00:00:50.200 --> 00:00:56.340
The wording suggests red without video, so success may reflect
text rather than visual understanding.
即使没有视频，措辞也强烈暗示红色；答对可能反映文本，而非视觉理解。

9
00:00:57.400 --> 00:01:03.140
A video may show the same answer, but its presence does not make
the question depend on visual evidence.
视频可能显示同一答案，但出现视频并不会使问题依赖视觉证据。

10
00:01:04.260 --> 00:01:09.800
Our first question leaves the order open, because either object
could move first in an otherwise plausible scene.
第一个问题未限定顺序，因为任一物体都可能先移动。

11
00:01:10.760 --> 00:01:16.760
That separates evidence required for an answer from evidence
merely present while the answer is already inferable.
这区分回答所必需的证据，与答案已可推断时仅仅可用的证据。

12
00:01:17.820 --> 00:01:24.160
The paper calls text-based success linguistic shortcutting,
motivating a video-removal test of answerability.
论文称文本驱动成功为语言捷径，因此先移除视频检查可回答性。

13
00:01:24.860 --> 00:01:30.500
The model receives only the question and options, with video
content and subtitles entirely withheld.
诊断实验仅提供问题和选项，完全隐藏视频内容与字幕。

14
00:01:31.480 --> 00:01:37.780
Correct answers define text-only answerable items for that
model; other questions are treated as visually grounded.
正确回答的题目对该评估模型归为纯文本可答；其余作为视觉依赖题。

15
00:01:38.840 --> 00:01:45.040
These labels describe an operational test, so the distinction
depends on which model performs the text-only assessment.
这些标签描述操作性测试，因此区别取决于执行纯文本评估的模型。

16
00:01:46.180 --> 00:01:53.620
Shortcuts include wording, world knowledge, implausible-option
elimination, and imagined content that happens to match.
论文识别措辞、外部知识、排除不合理选项及恰好匹配的想象内容等捷径。

17
00:01:54.660 --> 00:02:00.000
A higher score can include stronger language-only reasoning,
without equal gains in interpreting video.
更高基准得分因此可能包含更强纯语言推理，而不等于更强视频解读。

18
00:02:00.840 --> 00:02:06.620
Testing without visuals reveals this contribution, rather than
crediting every correct answer to visual comprehension.
去掉视觉输入让这种贡献可见，而不是默认每个正确答案都反映视觉理解。

19
00:02:07.260 --> 00:02:13.200
The paper summarizes roughly forty to sixty percent text-only
answerability on the examined benchmarks.
对所检查基准，论文概括约百分之四十至六十的题目仅靠文本线索可答。

20
00:02:13.800 --> 00:02:19.460
The exact proportion depends on the benchmark and assessing
model, not a universal set of video questions.
精确结果取决于基准与评估模型，不是所有视频问题的通用比例。

21
00:02:20.440 --> 00:02:28.520
Table one reports GPT-five text-only accuracy of forty-eight
point two on VideoMME, with video input withheld.
表一报告隐藏视频时 GPT-5 在 VideoMME 的纯文本准确率为四十八点二。

22
00:02:29.380 --> 00:02:36.340
For multiple-choice MMVU, the score is fifty-seven point one;
chance accuracy differs across benchmark designs.
在选择题 MMVU 上相应得分为五十七点一；各基准的随机水平不同。

23
00:02:37.460 --> 00:02:43.620
These correct-answer rates without video show why accuracy alone
cannot isolate the added value of watching.
这些是没有视频时的正确率，说明基准准确率本身无法分离观看的额外价值。

24
00:02:44.420 --> 00:02:50.160
Post-training data has similar shortcut questions, which can
reward language priors instead of visual reasoning.
问题也出现在后训练数据中，捷径题可能奖励语言先验而非要求视觉推理。

25
00:02:50.800 --> 00:02:57.700
VidGround curates Video-R1 post-training data, containing both
video and image question-answer examples.
VidGround 筛选 Video-R1 后训练池，其中同时含视频与图像问答样本。

26
00:02:58.439 --> 00:03:05.100
GPT-five mini sees questions and options; selection keeps those
it cannot correctly answer without visuals.
GPT-5-mini 只读取问题和选项，选择它没有视觉输入就不能答对的题目。

27
00:03:05.860 --> 00:03:12.100
The retained pool has one hundred eighty-one thousand seven
hundred ten examples, sixty-nine point one percent.
保留集合含十八万一千七百一十样本，占原池百分之六十九点一。

28
00:03:12.840 --> 00:03:18.680
The full pool has two hundred sixty-three thousand seventy-one
examples, mixing video and image questions.
完整比较池含二十六万三千零七十一样本；保留比例针对这个训练集。

29
00:03:19.600 --> 00:03:25.500
This training-data ratio differs from the benchmark diagnosis,
which measures answerability on separate datasets.
它不同于百分之四十至六十的基准诊断，后者对应其他数据集与评估模型。

30
00:03:26.480 --> 00:03:33.120
Text-only failure defines selection; it does not guarantee every
retained question has no possible linguistic shortcut.
纯文本失败是筛选规则，并不保证每个保留样本都没有任何可能的语言捷径。

31
00:03:33.653 --> 00:03:40.353
The approach combines data selection with RL post-training,
targeting examples that require visual evidence.
方法将数据筛选与强化学习后训练结合，针对需要视觉证据的样本。

32
00:03:41.253 --> 00:03:47.833
Qwen two point five VL seven B Instruct is trained with sixteen
uniformly sampled frames per video.
主配置后训练 Qwen2.5-VL-7B-Instruct，每视频均匀采样十六帧。

33
00:03:48.493 --> 00:03:54.893
Training uses seven hundred steps with the GRPO-based
optimization and temporal rewards stated in the source.
训练进行七百步，采用 GRPO 方案及论文说明的时序感知奖励组件。

34
00:03:55.993 --> 00:04:05.093
VideoMME, VideoMMMU, and multiple-choice MMVU are evaluated with
sixteen, thirty-two, or sixty-four video frames.
以十六、三十二或六十四帧，评估 VideoMME、VideoMMMU 和选择题 MMVU。

35
00:04:06.092 --> 00:04:11.693
Full benchmark and visually grounded subset averages use
separate populations, which must not be mixed.
完整基准均值与视觉依赖子集均值单独报告，因此不能混合两个评估群体。

36
00:04:12.533 --> 00:04:18.173
The ablation compares full-pool and selected-pool training,
isolating the effect of data composition.
数据消融比较完整池与选择池训练，明确后训练数据构成的贡献。

37
00:04:18.780 --> 00:04:26.040
At sixty-four frames, unfiltered-pool GRPO reports a fifty-three
point three Full average in Table three.
六十四帧评估时，表三报告完整池训练 GRPO 的 Full 均值为五十三点三。

38
00:04:27.160 --> 00:04:33.320
Training on the selected visually grounded pool reaches
fifty-nine point five on that same full-benchmark average.
在选出的视觉依赖池上训练，同一完整基准平均值达到五十九点五。

39
00:04:34.340 --> 00:04:41.160
The six-point-two-percentage-point difference belongs to this
model, data comparison, frame count, and population.
差距为六点二个百分点，限定于这个模型、数据比较、帧数和评估群体。

40
00:04:42.160 --> 00:04:48.380
The visually grounded subset instead compares forty-two point
nine with forty-seven point nine: five points.
视觉依赖子集比较不同：四十二点九对四十七点九，提升五点。

41
00:04:49.260 --> 00:04:55.860
Thus the reported up-to-six-point-two improvement is not the
gain for every frame setting, task, or subset.
因此报告的最多六点二提升，并非每个帧设置、任务或子集的收益。

42
00:04:56.760 --> 00:05:02.860
The result supports checking which examples require visual
evidence, rather than assuming more data must be better.
匹配结果支持检查哪些样本需要视觉证据，而不是认为更大训练池必然更好。

43
00:05:03.467 --> 00:05:09.807
Event order needs the displayed sequence, while a leading color
question can already suggest its answer.
回到两个运动物体：事件顺序需要显示序列，而引导性颜色题可能暗示答案。

44
00:05:10.887 --> 00:05:16.187
VidGround screens text-only answerability before selecting
examples to guide post-training.
VidGround 把区别转为数据筛选流程，先检查可回答性，再决定后训练样本。

45
00:05:17.007 --> 00:05:23.107
Selected examples provide visual-dependent supervision under the
filter, alongside an existing RL approach.
选择池提供操作性地需要视觉输入的监督，并与既有强化学习方法配对。

46
00:05:24.047 --> 00:05:30.867
Measured gains are configuration-specific; the deeper lesson
concerns the evidence that training examples reward.
报告收益对应具体基准设置，更深的启示在于训练样本奖励了何种证据。

47
00:05:31.907 --> 00:05:38.227
Evaluation should separate language-only success from the extra
contribution of watching the actual video.
评价应区分纯语言成功与观看额外贡献，而非把每个答案都归功于视频理解。

48
00:05:39.087 --> 00:05:44.647
For a model meant to watch before answering, the training
question should make the visual evidence matter to the answer.
对应先看再答的模型，训练问题应让视觉证据对答案真正重要。