Sabitlenmiş Tweet
碧落空歌 GrumpyAndFoxy
187 posts
碧落空歌 GrumpyAndFoxy retweetledi
碧落空歌 GrumpyAndFoxy retweetledi
We've published a paper that explains our views on AI competition between the US and China.
The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: anthropic.com/research/2028-…
English
几句话，男女都该看看。
45岁之后，人生的残酷程度会乘以10倍。那之后的形象，基本靠人民币维持，不是靠你多能喝酒多能侃。
性是肉体生活，遵循快乐原则。爱情是精神生活，遵循理想原则。婚姻是社会生活，遵循现实原则。三件事，从来都不是一回事。
男女之间那点意思，通常从"不好意思"开始，到"真没意思"结束。
当别人不再需要你，学会收回热情，礼貌退场。你可以躲在被窝里哭，可以喝到吐，但不能拿起手机发不该发的消息。一扇不愿意开的门，一直敲是没有礼貌的。
结不结婚都会后悔。巷子里的猫很自由，但没有归宿。围墙里的狗有归宿，但终身低头。怎么选都有遗憾，但记住一件事——谋爱之前先谋生，爱人之前先爱自己。没有经济能力，是万劫不复。
恋爱最好找同类，婚姻最好找互补。恋爱是情感组合，开心就够了。婚姻是价值组合，能帮彼此实现价值才稳固。
最好的伴侣，是你人生战场上的盟友，不是保姆。真正稀缺的不是有人给你买早餐送夜宵，是对方的眼界、情绪稳定和控制局面的能力。
别人对你的态度，由两件事决定——你当下的结果，和你贡献的好处。
要调整吗？
中文
@skdh One of these anomalies will turn out to be real , but we need new theories that show us where to look x.com/Davidmdrpi/sta…
David5D@Davidmdrpi
Physicist John A. Wheeler gave us “It from Bit”: the idea that physical reality may ultimately arise from information. #BAM is exploring the reverse possibility: “Bit from It.” Not information first, but geometry first. Not quantum rules imposed on spacetime, but quantum-like discreteness emerging because a closed spacetime geometry only permits certain self-consistent resonances. Notes our universe can hold. In closed ringing spherical universe where particle pairs are wormholes and all wave particle interactions are antipodal. The closure-ledger work now shows a coherent chain: 2pi action ledger to a self-selected outer radius to gamma, transport, and resistance as geometric closure-quantum quantities to an inner cutoff epsilon to the Compton bridge leaving only the electron scale as the remaining physical anchor. The newest throat dynamics pass sharpened the picture further. Local boundary condition swaps failed. Smooth finite-thickness throat barriers also failed. They introduced extra arbitrary parameters instead of explaining the boundary. But the reflection-phase analysis found something deeper: the successful hardwall cutoff behaves like an effective representation of a nonlocal Bohr-Sommerfeld throat phase. At the lepton-selected geometry, the lowest radial mode satisfies approximately: omega times L is about gamma over 2pi, close to 7/2. Then uniform WKB decomposes the correction into explicit Tangherlini radial pieces. That does not yet prove the gamma identity analytically, but it shows the missing phase is not random. It lives in the radial throat structure. So the story is no longer “choose a cutoff and fit the lepton ladder.” It is becoming: closed geometry imposes action counts, the lepton spectrum selects the cavity, the throat boundary encodes a nonlocal phase, and the quantum ledger emerges from classical geometry. Bulk Antipodal Mechanics BAM is not yet a derivation of all Quantum Mechanics from General Relativity. But it is exactly the direction Einstein and Wheeler might have hoped existed: matter, phase, and quantum discreteness appearing as properties of spacetime structure itself. It from Bit may have been backwards. The breathtaking possibility is Bit from It!
English
For the first time in years, particle physics has a new anomaly, or rather, the return of a previous one that just won't go away. What could it be? I have a brief summary
English
You are looking at NOBEL PRIZE WINNING work.
These are the stars orbiting the supermassive black hole at the heart of our galaxy. Here is everything you need to know about it:
English
"Folks, there's an ongoing battle in the world between autocracy and democracy. Xi Jinping,..who I've talked... the fact of the matter is, he just is straightforward about it. He says that democracies cannot be sustained in the 21st century. Not a joke. They cannot be sustained."
De Moin@DeMoinug
@RJDAIGOGO 拜登关于习近平的话，我找到了原文出处 presidency.ucsb.edu/documents/rema…
English
Easting the west :P
the cover image of my book was taken by Wu Guoyong for his 中国白宫 project, where he photographed 100+ capitol hill inspired buildings in China. this one is located in a filming studio in Hebei.
English
@DestinyLinker 2025第16届全球算命师大赛, 比赛试题：
zhuanlan.zhihu.com/p/191979431021…
中文
Tianfu Agent 在全球算命师大赛上跑到 50% 截尾准确率（人类 Top-20 选手平均 53.5%）
比赛 3069 名参赛者
人类 Top-20 选手平均 53.5%
最强通用大模型基线（Claude Opus 4.6）40%，这中间差了 10 个百分点
1️⃣ 一句话讲清楚
一个为命理术数专门设计的 agent 系统，在中国传统文化领域里，第一次真正贴近顶尖人类选手的水平
2️⃣ 它是什么？
给 LLM 造了一整套命理专用工具环境
200+ 原子工具 / Agentic 端到端推理
让 AI 真正学会怎么「做命理」 跳出「把命盘数据塞进 Prompt 让通用大模型硬猜」这条老路子
3️⃣ 以前的解决方案
「排盘数据 + 通用大模型」 听起来够用了，实际上有三个结构性硬伤：
1）衍生数据会组合爆炸
大限 / 流年 / 飞宫 层层展开 没法穷举塞进 Prompt
2）空间关系序列化造成幻觉
三方四正 / 能量流通 全是拓扑结构 翻译成文字就面目全非
3）推理链越长越飘
每一步都依赖上一步 错误逐步放大
专业训练语料几乎为零 模型压根不懂这些规则
4️⃣ Tianfu Agent 的思路换了一套
第一 确定性优先
200+ 专用原子工具
排盘 / 飞宫 / 取用神推演 全部精确计算
模型不用「回忆」知识
第二 推理规则工具化
行业内部的推理技法 也写成可调用函数
该用哪条 / 什么时候用 模型按需精准触发
绕开了「让模型记住并遵守专业规则」这条永远跑不通的路
第三 量化「直觉」
从工具输出量化指标 / Sub-Agent 的自评 / 多流派合参的调和
层层量化 模拟人类专家的隐式判断直觉
5️⃣ 技术报告
1）技术报告 + 原始答案：destinylinker.github.io/MingLi-Bench/
2）Benchmark 数据 + 评测代码：github.com/DestinyLinker/…
做 agent 或者做传统文化 AI 的，可以麻烦仓库点颗星支持一下🌟
下一条把「200+ 原子工具」这套工具栈具体长什么样拆出来
中文
全球算命师大赛(香港）
hkjfma.org/2025/06/2025%e…
DestinyLinker@DestinyLinker
Tianfu Agent 在全球算命师大赛上跑到 50% 截尾准确率（人类 Top-20 选手平均 53.5%） 比赛 3069 名参赛者 人类 Top-20 选手平均 53.5% 最强通用大模型基线（Claude Opus 4.6）40%，这中间差了 10 个百分点 1️⃣ 一句话讲清楚 一个为命理术数专门设计的 agent 系统，在中国传统文化领域里，第一次真正贴近顶尖人类选手的水平 2️⃣ 它是什么？ 给 LLM 造了一整套命理专用工具环境 200+ 原子工具 / Agentic 端到端推理 让 AI 真正学会怎么「做命理」 跳出「把命盘数据塞进 Prompt 让通用大模型硬猜」这条老路子 3️⃣ 以前的解决方案 「排盘数据 + 通用大模型」 听起来够用了，实际上有三个结构性硬伤： 1）衍生数据会组合爆炸 大限 / 流年 / 飞宫 层层展开 没法穷举塞进 Prompt 2）空间关系序列化造成幻觉 三方四正 / 能量流通 全是拓扑结构 翻译成文字就面目全非 3）推理链越长越飘 每一步都依赖上一步 错误逐步放大 专业训练语料几乎为零 模型压根不懂这些规则 4️⃣ Tianfu Agent 的思路换了一套 第一 确定性优先 200+ 专用原子工具 排盘 / 飞宫 / 取用神推演 全部精确计算 模型不用「回忆」知识 第二 推理规则工具化 行业内部的推理技法 也写成可调用函数 该用哪条 / 什么时候用 模型按需精准触发 绕开了「让模型记住并遵守专业规则」这条永远跑不通的路 第三 量化「直觉」 从工具输出量化指标 / Sub-Agent 的自评 / 多流派合参的调和 层层量化 模拟人类专家的隐式判断直觉 5️⃣ 技术报告 1）技术报告 + 原始答案：destinylinker.github.io/MingLi-Bench/ 2）Benchmark 数据 + 评测代码：github.com/DestinyLinker/… 做 agent 或者做传统文化 AI 的，可以麻烦仓库点颗星支持一下🌟 下一条把「200+ 原子工具」这套工具栈具体长什么样拆出来
中文
When people say “from the photon's frame of reference, no time passes” or “the universe is squashed to zero length”, these are informal ideas, not actual physics. In reality, a photon is not a valid observer.
A better way to think about this is in terms of motion through spacetime.
In relativity, everything moves through 4D spacetime (3 space + 1 time) with a total “speed” equal to c. What changes is how this is divided between space and time:
1. An object at rest in space moves entirely through the time dimension at speed c.
2. An object moving fast through space trades some of its motion through time for motion through space. Its total spacetime “speed” is still c, but now it's split between time and space.
3. A photon moves entirely through space at speed c, with no motion through time.
A photon uses all of its spacetime motion for spatial movement, leaving nothing for time. That’s why, informally we say “no time passes for a photon.”
Mathematically, this shows up in the spacetime interval:
s² = (ct)² − x² − y² − z²
For massive objects, this value is positive (timelike). For light, it is exactly zero (lightlike or null), which is why light follows what’s called a null path.
We describe photons from outside, using a valid frame. We say things like “the photon moves at c in space and has zero proper time along its path.” This is well-defined.
What’s not well-defined is trying to sit on the photon and ask what it experiences. To experience anything, proper time must pass, and proper time is zero along a photon’s path.
So instead of asking what a photon experiences, it makes more sense to look at how spacetime behaves along its path.
English
碧落空歌 GrumpyAndFoxy retweetledi
Google trained an AI to predict your neighbourhood's income by counting the coffee shops, bus stops, and high-rises on a map. Nobody told it what income was.
The model is called S2Vec, published this month by Google Research as part of their Earth AI initiative. It takes the built environment (every building, road, park, and business in an area) and converts it into a layered image. Three coffee shops and one park in a grid cell become pixel values. The AI then reads that image the same way a computer vision model reads a photograph.
The training method is the part that matters. S2Vec uses masked autoencoding: you show the model a patch of a city with chunks missing, and it learns to fill in the gaps. Show it a cluster of high-rise apartments next to a subway station, mask out a section, and it predicts a grocery store belongs there.
Do that millions of times across the globe and the model learns the deep spatial grammar of how cities organise themselves. No human ever labels a region as "financial district" or "suburban residential." The model figures out those groupings on its own from the geometry of what's built where.
The output is an embedding, a string of numbers that acts as a mathematical fingerprint for any location on Earth. Feed those embeddings into a prediction task and S2Vec can estimate population density, median income, and carbon emissions for regions it has never seen before.
On zero-shot geographic extrapolation (predicting for regions entirely absent from training data) S2Vec was typically the best-performing individual model.
It matched or beat satellite imagery baselines like RS-MaMMUT and outperformed GEOCLIP on socioeconomic prediction. The best results came from combining S2Vec with satellite image embeddings. Built environment data alone couldn't capture vegetation, terrain, or transportation patterns well enough for environmental tasks like tree cover and elevation. But fused together, the two modalities outperformed everything else.
The standard approach to geospatial ML has been hand-crafting indicators for every new problem. Predicting air quality meant building a bespoke feature set. Estimating housing prices meant building another one. S2Vec replaces that with a single general-purpose representation that transfers across tasks.
The training data is map features, not satellite pixels.
That distinction is pretty important to understand. It means: map data updates faster, costs less to process, and covers built infrastructure at a resolution satellite imagery can't always match.
A satellite sees rooftops. S2Vec knows there are three cafes, a pharmacy, and a bus stop underneath them.
Google's broader Earth AI pipeline now has three foundation models working in parallel.
1. PDFM for population dynamics.
2. RS-MaMMUT for satellite imagery.
3. S2Vec for the built environment.
Stack them and you get a system that can read a neighbourhood the way a local understands it.
English
