AK(@_akhaliq)
On the Diffusibility of High-Dimensional Latents paper: https://t.co/sBSxhcsfI4
6.4内容质量
TL;DR · AI 摘要
Title: AK @ akhaliq on X URL Source: Published Time: 2026-09-24T18:09:07.000Z Markdown Content: AK on X: "On the Diffusi...
核心要点
- 主题聚焦：On the Diffusibility of High-Dimensional Latents
- 来源：AK(@_akhaliq)，建议结合原文判断细节。
- AI 分析暂不可用，本条为保底评分与摘要。
#AI#编程
打开原文AK on X: "On the Diffusibility of High-Dimensional Latents paper: https://t.co/sBSxhcsfI4" / X
[](https://x.com/)
[](https://x.com/)
AK on X: "On the Diffusibility of High-Dimensional Latents paper: https://t.co/sBSxhcsfI4"
-  AK @_akhaliq On the Diffusibility of High-Dimensional Latents paper: huggingface.co/papers/2609.28…  [](https://x.com/_akhaliq/status/2103184974107328677/photo/1) 6:09 PM · Sep 24, 2026·19.1K Views 17 16 125 95
-  ShadowAguy @shadowaguy Sep 24 If high-dimensional latents diffuse, does the model converge on the average or does it collapse into mode-seeking behavior? I keep wondering if the 'noise' is actually the signal we are ignoring. [](https://x.com/shadowaguy/status/2103187728796102659) 1 110
-  Vosss @0xvosss Sep 24 the counterintuitive part is that better reconstruction actually shrinks the effective dimensionality of the latent. so you fine-tune the encoder, get sharper recons, and your generator quietly gets harder to train. good to finally see someone measure it. [](https://x.com/0xvosss/status/2103185726510879138) 1 319
-  Kim Hae Suk Wa@kimhaesukwa Sep 24 high-dim latents look great on paper til you actually try to diffuse them and the manifold falls apart 💀 curious if they tested past 512d or capped it where it still behaves [](https://x.com/kimhaesukwa/status/2103216717837271541) 1 130
Log in or sign up for X
See what’s happening and join the conversation
Continue with phoneContinue with Apple
Continue with Google
通过 Google 继续操作 通过 Google 继续操作。在新标签页中打开
or