Julius Horsthuis - Fractalicious (2014)
https://www.youtube.com/watch?v=vsQBR68IiNA
https://www.youtube.com/watch?v=vsQBR68IiNA
YouTube
Fractalicious
I published my first Fractal on the 31st of December 2013. After that, 2014 has been a very Fractalicious year for me. Thanks to a very supportive community, I’ve produced some 20 little fractal clips that total more than an hour of running time. Currently…
&'a ::rynco::UntitledChannel
Julius Horsthuis - Fractalicious (2014) https://www.youtube.com/watch?v=vsQBR68IiNA
Vessels - The Sky Was Pink (2012)
https://www.youtube.com/watch?v=ggLh9P9x24Q
https://www.youtube.com/watch?v=ggLh9P9x24Q
YouTube
Vessels - The Sky Was Pink (official video)
Vessels - The Sky Was Pink (2012)
A cover of James Holden's remix of Nathan Fake's classic.
Written by Nathan Fake (Chrysalis Music)
Video directed by Morgan Beringer
http://www.morganberinger.com
Free download here: http://soundcloud.com/vessels/vessels…
A cover of James Holden's remix of Nathan Fake's classic.
Written by Nathan Fake (Chrysalis Music)
Video directed by Morgan Beringer
http://www.morganberinger.com
Free download here: http://soundcloud.com/vessels/vessels…
Forwarded from Hacker News (yahnc_bot)
Watching AI drive Microsoft employees insane https://old.reddit.com/r/ExperiencedDevs/comments/1krttqo/my_new_hobby_watching_ai_slowly_drive_microsoft/
Reddit
From the ExperiencedDevs community on Reddit: My new hobby: watching AI slowly drive Microsoft employees insane
Explore this post and more from the ExperiencedDevs community
🤡5🤝1
Forwarded from Hacker News
Magic of software; what makes a good engineer also makes a good engineering org (❄️ Score: 153+ in 2 days)
Link: https://readhacker.news/s/6uH99
Comments: https://readhacker.news/c/6uH99
Link: https://readhacker.news/s/6uH99
Comments: https://readhacker.news/c/6uH99
Moxie Marlinspike
The magic of software; or, what makes a good engineer also makes a good engineering organization
The people who create software generally refer to themselves as software engineers, and yet if they graduate from university, it is typically with a degree in computer science. That has always felt a little strange to me, because science and engineering are…
.NET 的全新低延时高吞吐自适应 GC - Satori GC
https://zhuanlan.zhihu.com/p/1908681293006354245
https://zhuanlan.zhihu.com/p/1908681293006354245
利用嵌入的通用几何结构
Harnessing the Universal Geometry of Embeddings
Rishi Jha, Collin Zhang, Vitaly Shmatikov, John X. Morris
https://arxiv.org/abs/2505.12540
我们首次提出无需配对数据、编码器或预定义匹配集即可实现文本嵌入在不同向量空间转换的方法。这种无监督方法可将任意嵌入映射至通用潜在表示（即柏拉图表示假设提出的通用语义结构），并实现双向转换。我们的转换在架构参数量级不同、训练数据集各异的模型对之间实现了高余弦相似度。
这种保持几何特性转换未知嵌入向量的能力，对向量数据库安全具有重大影响。攻击者仅凭嵌入向量即可提取底层文档的敏感信息，足以进行内容分类和属性推断。
作者的讲解帖子
https://fixupx.com/jxmnop/status/1925224612872233081
https://fixupx.com/rishi_d_jha/status/1925212069168910340
Harnessing the Universal Geometry of Embeddings
Rishi Jha, Collin Zhang, Vitaly Shmatikov, John X. Morris
https://arxiv.org/abs/2505.12540
我们首次提出无需配对数据、编码器或预定义匹配集即可实现文本嵌入在不同向量空间转换的方法。这种无监督方法可将任意嵌入映射至通用潜在表示（即柏拉图表示假设提出的通用语义结构），并实现双向转换。我们的转换在架构参数量级不同、训练数据集各异的模型对之间实现了高余弦相似度。
这种保持几何特性转换未知嵌入向量的能力，对向量数据库安全具有重大影响。攻击者仅凭嵌入向量即可提取底层文档的敏感信息，足以进行内容分类和属性推断。
作者的讲解帖子
https://fixupx.com/jxmnop/status/1925224612872233081
https://fixupx.com/rishi_d_jha/status/1925212069168910340
arXiv.org
Harnessing the Universal Geometry of Embeddings
We introduce the first method for translating text embeddings from one vector space to another without any paired data, encoders, or predefined sets of matches. Our unsupervised approach...
🤯8🤔1🤩1
🔥11🤩1
YDPFa:
#ガールズバンドクライ
#になもも
#100日チャレンジ
hug (68/100)
https://twitter.com/YDPFALION/status/1925190773588611459
#ガールズバンドクライ
#になもも
#100日チャレンジ
hug (68/100)
https://twitter.com/YDPFALION/status/1925190773588611459
🍓9🥰2💋1
Diffusion is Spectral Autoregression
https://sander.ai/2024/09/02/spectral-autoregression.html
https://sander.ai/2024/09/02/spectral-autoregression.html
Sander Dieleman
Diffusion is spectral autoregression
A deep dive into spectral analysis of diffusion models of images, revealing how they implicitly perform a form of autoregression in the frequency domain.
大型语言模型不具备类人工作记忆能力
LLMs Do Not Have Human-Like Working Memory
Jen-tse Huang, Kaiser Sun, Wenxuan Wang, Mark Dredze
https://arxiv.org/abs/2505.10571
人类工作记忆是一种主动的认知系统，不仅能实现信息的临时存储，还能支持信息的加工与运用。缺乏这种机制可能导致个体产生不真实的对话内容、出现自相矛盾，以及在需要心理推理的任务中表现欠佳。本文通过三项实验研究（1）数字猜测游戏（2）是非问答游戏（3）数学魔术，论证当前主流大语言模型尚未具备这种类人的认知能力，这将成为实现通用人工智能的重要障碍。在多个模型系列上的实验结果表明，现有模型在这些场景下均未能展现出类人的认知行为。通过揭示这一局限性，我们呼吁学界进一步探索如何提升大语言模型的工作记忆能力。
LLMs Do Not Have Human-Like Working Memory
Jen-tse Huang, Kaiser Sun, Wenxuan Wang, Mark Dredze
https://arxiv.org/abs/2505.10571
人类工作记忆是一种主动的认知系统，不仅能实现信息的临时存储，还能支持信息的加工与运用。缺乏这种机制可能导致个体产生不真实的对话内容、出现自相矛盾，以及在需要心理推理的任务中表现欠佳。本文通过三项实验研究（1）数字猜测游戏（2）是非问答游戏（3）数学魔术，论证当前主流大语言模型尚未具备这种类人的认知能力，这将成为实现通用人工智能的重要障碍。在多个模型系列上的实验结果表明，现有模型在这些场景下均未能展现出类人的认知行为。通过揭示这一局限性，我们呼吁学界进一步探索如何提升大语言模型的工作记忆能力。
arXiv.org
On the Failure of Latent State Persistence in Large Language Models
While Large Language Models (LLMs) excel in reasoning, whether they can sustain persistent latent states remains under-explored. The capacity to maintain and manipulate unexpressed, internal...
🤷♀10👀3👍1🤮1💩1🤡1