`包裹,长文用`
`/`
`划分小节。 3. 列表内容用`
- `,引用用`
`,代码用`
`,重要句子可用``。
4. 保留原文所有信息,不得增删改内容。
5. 不要输出``/`/`等文档级标签。
分析正文内容:
正文是一段学术新闻/研究简报,包含多个自然段落。
段落1:Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question.
段落2:We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect.
段落3:The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity.
虽然正文不算特别长,但可以加一个小标题,或者直接用段落。为了体现``或`
`,可以加一个合适的小标题,比如“研究背景与方法”、“实验结果与发现”等,但规则说“不得增删改内容”,如果加小标题算不算增改内容?
规则说“保留原文所有信息,不得增删改内容”。如果加小标题,属于增加了原文没有的文字。所以最好不加小标题,或者把小标题用原文的词语?不,安全起见,不添加任何原文没有的文字。那`
/
`就可以不用,或者如果必须用,可以用原文的句子?不,规则说“长文用
/
划分小节”,不是强制必须用,而是“用”。如果不用也可以。
不过,为了严格不增加内容,我可以把第一句或者某些句子作为标题?不,直接全部用`
`包裹即可。或者把段落里的某句话用``加粗,体现“重要句子可用``”。 让我们仔细看原文: 段落1: Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question. 段落2: We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect. 段落3: The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity. 为了使用``,可以把一些关键结论加粗。 例如:
Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question.
We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect.
The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity.
检查是否增删改内容:
加粗只是格式变化,没有改变文字。
没有添加小标题,所以没有违反“不得增删改内容”。
没有列表、引用、代码,所以不需要那些标签。
再检查一遍文本是否完全一致:
Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question.
We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect.
The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity.
完全一致。
输出格式要求:只输出HTML片段本身,不要Markdown围栏、不要解释。不要输出//等文档级标签。
构建最终输出:
Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question.
`,代码用``,重要句子可用``。 4. 保留原文所有信息,不得增删改内容。 5. 不要输出``/`/`等文档级标签。 分析正文内容: 正文是一段学术新闻/研究简报,包含多个自然段落。 段落1:Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question. 段落2:We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect. 段落3:The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity. 虽然正文不算特别长,但可以加一个小标题,或者直接用段落。为了体现``或`
`,可以加一个合适的小标题,比如“研究背景与方法”、“实验结果与发现”等,但规则说“不得增删改内容”,如果加小标题算不算增改内容? 规则说“保留原文所有信息,不得增删改内容”。如果加小标题,属于增加了原文没有的文字。所以最好不加小标题,或者把小标题用原文的词语?不,安全起见,不添加任何原文没有的文字。那`
/
`就可以不用,或者如果必须用,可以用原文的句子?不,规则说“长文用
/
划分小节”,不是强制必须用,而是“用”。如果不用也可以。 不过,为了严格不增加内容,我可以把第一句或者某些句子作为标题?不,直接全部用`
`包裹即可。或者把段落里的某句话用``加粗,体现“重要句子可用``”。 让我们仔细看原文: 段落1: Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question. 段落2: We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect. 段落3: The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity. 为了使用``,可以把一些关键结论加粗。 例如:
Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question.
We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect.
The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity.
检查是否增删改内容: 加粗只是格式变化,没有改变文字。 没有添加小标题,所以没有违反“不得增删改内容”。 没有列表、引用、代码,所以不需要那些标签。 再检查一遍文本是否完全一致: Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question. We set out to find hidden physical knowledge in kinematic text-to-motion diffusion models, where the physics of every intermediate estimate can be computed exactly, and recorded every layer along the sampling trajectories of MDM (two checkpoints) and MotionDiffuse. Against the usual baselines the layers do look ahead of the model's own estimate on balance and flight. With every readout tuned the same way, the estimate closes the gap: for balance and flight, adding a layer gains at most 0.01, a size that a positive control shows we would detect. The fall rate of jumps is the one clear exception. Early in denoising, MDM's layers add about 0.08 in $R^2$ (up to 0.11 in a refit) that the estimate does not yet show, and the gain fades as sampling proceeds; on MotionDiffuse it is small and not robust. The estimate also flags floating samples without training, but aborting them saves under a tenth of the compute. In a video model, plausibility that linear probes miss in Wan 2.1's clean latent becomes readable after one fixed nonlinearity. 完全一致。 输出格式要求:只输出HTML片段本身,不要Markdown围栏、不要解释。不要输出//等文档级标签。 构建最终输出:Recent studies report that video diffusion models "know more than they show": physical properties decode better from their activations than from their inputs or outputs. But a diffusion model computes its clean-sample estimate from those same activations, so such a comparison measures how easily a readout extracts the information. Whether the model hides anything is a separate question.