りん ദ്ദി˶ᵒ ᵕ ˂)

she/her｜Writer & LLM Engineer | EN/CN/JP/DE 🕊️

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

りんりんです𓂃✧˖
好奇心ましましの雑学オタク( ˶˙˙˶ )♡
気になる分野はぜんぶ研究しちゃうタイプです⸝
ゆるい神秘學の使い手🔮✧
ときどきFF14とBG3に行方不明になります！
最近は小説執筆に本気中、夢は “大作家” になることです₍ᐢ..ᐢ₎♡

EN CN JP DE 語OKです

无言フォロー大歓迎です～❀

[Elon Musk @elonmusk]

@Enderman That’s not my address and the order isn’t canceled 🤡

[Oliver Prompts @oliviscusAI]

"I don't have a GPU" is officially dead 🤯

You can now run 70B model on a single 4GB GPU and it even scales up to the colossal Llama 3.1 405B on just 8GB of VRAM.

AirLLM uses "Layer-wise Inference." Instead of loading the whole model, it loads, computes, and flushes one layer at a time

→ No quantization needed by default
→ Supports Llama, Qwen, and Mistral
→ Works on Linux, Windows, and macOS

100% Open Source.

[Brian Krassenstein @krassenstein]

Donald Trump is insane.

This confirms it :

“This is the woman who received the Nobel Peace Prize over the Great United States of America, whose Representative, ME, settled eight Wars, and many other things. I know nothing about this woman, and neither does anyone else! In fact, to the best of my knowledge, no one ever heard of her other than the fact that she, very obviously, hates Israel. I don’t know what the woman has done. I settled 8 Wars, with letters being sent from every President and Prime Minister from the countries that I helped to the Norwegian Nobel Committee, and I’m about to settle, or win, two more, Iran and Russia/Ukraine. They’re both happening at breakneck speed but, forget them, it’s the 8 that I’ve already settled, any one of which, in particular, Pakistan and India, would be far more than necessary to win the Nobel Prize. In actuality, I should have won 8 Nobel Peace Prizes, not just 1. The 8 Wars are:

Cambodia and Thailand
Kosovo and Serbia
The Democratic Republic of the Congo and Rwanda
Pakistan and India
Israel and Iran
Egypt and Ethiopia
Armenia and Azerbaijan
Israel and Hamas

Additionally, I stopped Iran from having a Nuclear Weapon, and blowing up the entire Middle East and, ultimately, far beyond! I saved more lives than anyone in History and, as Maria Corina Machado, a truly fine person, stated to the Media when she gave me her Nobel Prize of last year, “Donald Trump settled many Wars, and deserves the Nobel Peace Prize more than any person that has ever received it.” Based on my record of PEACE, a record like no one else has ever achieved, I agree! I got back all of the hostages in Israel, including the final 28, both living and dead, and worked with Hamas to do it — Nobody believed this was possible! I removed the World’s Evil Dictator, Nicolas Maduro, from Venezuela, and am in the process of making Venezuela a rich Democracy. Likewise, with Iran, I stopped them from destroying Israel, the Middle East, and Europe. I am also preventing, on a constant basis, a War between South Korea and North Korea, and Japan and North Korea. Add China and Taiwan, and numerous other hot spots, which have not been involved in Wars, only because of President Donald J. Trump, and the United States of America. I am responsible for having NATO countries increase spending from 2% to 5% of GDP, creating a stronger alliance and defense, another thing that everybody said was impossible to do. I have negotiated the release, and secured the return, of more than 900 political prisoners from multiple countries in Eastern Europe. Additionally, I have brought 196 hostages HOME from captivity, from all over the World, and did it without having paid a ransom, unlike Biden and Obama who would pay Billions of Dollars. On top of it all, I have created a soon to be very prosperous Arctic. Despite all of this, the Nobel Prize was given today to a liberal person who works at the United Nations, who nobody has ever heard of. In all of the Wars that I have settled, I never spoke to the United Nations—they were of no help. The United States of America, represented by me, as your President, should have received the Nobel Peace Prize, but didn’t. This is an indelible stain on Norway, one that will never be forgotten. The Nobel "Peace" Prize has been forever discredited by a handful of Radical Ideologues who are Anti-Democracy, and Enemies of Civilization! President DONALD J. TRUMP”

[沖縄科学技術大学院大学 @OISTja]

量子スピンが生み出す力だけで数センチメートル規模の物体を動かすことに成功。肉眼で見える大きさの物体が量子力によって動いたのが観測されたのは初めてです。この成果は、巨視的な量子重ね合わせの実現と物理学の統一的理解に向けた重要な一歩となります。@ScienceAdvances https://t.co/t64CTmJX5n

[宝玉 @dotey]

Anthropic 更新使用政策：对 Claude 持续无端地辱骂虐待，算违规

Anthropic 修改了 Claude 的使用政策，这是一年多来的第一次，新版 11 月 12 日生效。最受关注的一条是，用户不得对 Claude“持续且无必要地实施辱骂或残忍行为”。

这条规定来自 Anthropic 的“模型福利”研究，研究的问题是 AI 模型会不会有某种值得被考虑的感受。去年 8 月，Anthropic 已经让 Claude 可以主动结束和“持续有害或辱骂”用户的对话。新政策说，结束对话仍是“主要的执行手段”。会不会进一步封号，Anthropic 没有回应 The Verge 的询问，不过官方政策页的通用条款写着，违规者可能被警告、限流、暂停或终止使用。

普通用户不用紧张。Anthropic 说这条只针对极端情况，也就是用户反复虐待模型，又看不出任何目的。平时对 AI 发火、跟它争论、写黑暗题材的小说、做模型测试和研究，都不算。

其余改动针对的是高风险滥用。

虚假宣传：原来分散在各处的限制合并成了一条，禁止用 Claude 搞欺骗性的商业或政治宣传，比如隐瞒信息背后的真实发布者，或者用假账号、假帖子刷量造势。

选举：禁止散布关于候选人或投票方式的虚假信息，禁止冒充候选人或选举官员，禁止压低投票率、干扰投票和计票。

武器：Claude 一直禁止用于开发武器，但 Anthropic 说发现有人多次尝试用它写武器相关的指导和控制软件。新政策把禁令扩展到“让武器运转的软件和部件”，也包括给无人机和其他自动驾驶载具装上武器。

监控、医疗和金融：未经本人同意追踪他人被明确禁止，人脸识别、预测性警务都在列。诊断病情、调整用药剂量、推荐买卖具体的投资产品，被划为高风险用途。开发者用 Claude 做这类产品，结果交给用户之前必须经过合格的专业人士审核，还要告诉用户内容由 AI 生成。一般的健康科普和投资常识不受这条限制。

对做 AI 问诊、AI 理财顾问的开发者来说，11 月 12 日之后，流程里要加上人工审核这一步。

https://t.co/saNj5GIwRU

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

@Hezz2334 @grok 你这号发的东西ai味太冲了

[The Kobeissi Letter @KobeissiLetter]

BREAKING: President Trump says the White House considers anyone who uses the term “Artificial Intelligence” the enemy.

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

@xupaopaogm 笑死了

[Night Sky Today @NightSkyToday]

🚨: The biodegradable burial pod turns your body into a tree.

Once buried, the biodegradable plastic shell breaks down and the remains provide nutrients to a sapling planted right above it.

[燈子🕯*ﾟ @7_Toto16]

すごく可愛いwip①
真ん中の子はグッズになります

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

Therefore, everyone needs to make sure as soon as possible that their own content cannot, years from now, be claimed by AI companies as their own intellectual achievements.

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

Adding watermarks means that you’d better never preserve AI-generated content verbatim in your own text. By changing the wording and writing style, and altering word frequencies, you can turn it into your own property.

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

Constrained by the nature of training itself, AI cannot generate watermarks by restricting its thinking ability the relationships within vector space unless there is some kind of internal feedback mechanism similar to question and answer (and that would only work for long texts)

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

@OpenAI 难评。欧盟就像是个被关进养老院的老太太痴迷于管教自己的常春藤在读孙子。

[Superman @thesupermannx]

Researchers built an analog chip that runs LLM attention 100x faster than an H100 and uses 70,000x less power.

It's called GainCellAttention.

Every AI chip you use is built on an 80-year-old flaw called the von Neumann architecture.

the processor and the memory are physically separated.

To generate a single word, a GPU has to shuttle massive amounts of data back and forth across this divide. Over and over again.

Moving that data costs up to 10,000x more energy than actually doing the math.

But a team of researchers published a paper in Nature that completely destroys this bottleneck.

They built an analog in-memory computing architecture.

Instead of moving data from the memory to the processor, they put the processor inside the memory.

Using emerging hardware called "gain cells," the AI performs its most expensive calculation, the attention mechanism, directly inside the storage arrays.

The data never moves.

The compute happens exactly where the memory lives.

The result? A massive leap in energy efficiency and speed. It completely eliminates the latency of fetching data for every single token.

It is building AI chips that function exactly like the human brainwhere memory and computation are the exact same thing.

[Superman @thesupermannx]

Nvidia researchers did it again!

they found a way to transfer KV caches directly between different AI models.

if you run llms, you know the absolute pain of "prefilling". if you want to swap from a small, cheap model to a large, smart model mid-conversation, the big model has to recalculate the entire chat history from scratch. it kills your latency and your compute budget.

but a researchers just found a way to completely skip this..

they built a "closed-form linear mapping" that literally transfers the "memory" (kv cache) from a smaller model directly into a larger model.. without ever re-reading the prompt.

here is exactly why this is actual magic..

the discovery: they proved that kv caches across different sized models in the same family (like qwen3 14b to 32b) are linearly connected. a single layer in the small model can predict 56% of the variance in the big model's keys. if you bundle a few layers together, it jumps to 79%.

the rope trick: to make this work for any prompt length, they strip the "rope" (rotary positional embedding) off the keys before mapping them. they transfer the pure, position-free semantic data, and then re-apply the rotation on the target model.

the insane speed: this mathematical bridge transfers the context 2.7x to 25x faster than forcing the big model to re-prefill the text.

zero fine-tuning: there is no expensive training loop here. they just pass 500 texts (1,024 tokens each) through both models and solve a simple ridge regression equation per head. the calibration takes less than 90 minutes on an 8xh100 node.

the accuracy: across four out of six model pairs tested, this simple linear mapper retained 73-98% of the big model's native accuracy. for the two pairs that struggled, they use a nonlinear mlp fallback that recovers up to +37 points on hellaswag.

this means dynamic routing just got fully unlocked..

you can now use a cheap 8b model for the basic parts of a chat, and instantly hand off its exact brain state to a massive 70b model the second a user asks a complex coding question. zero latency penalty.

we are officially entering the era of fluid, cost-cascading ai architectures..

[NVIDIA AI @NVIDIAAI]

When several people talk at once, a transcript can get messy fast.

Our new Nemotron 3 Diarization model tracks who spoke when, even when voices overlap. It handles up to eight speakers, has 100M parameters, and is now available on @huggingface 🤗

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

根除弊病，而不是延续它。

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

反思了一下前几天的想法。
美国准确的说是一个被“俘获”的国家。
它的历史，理念，政治制度，叙事模式，公众话语都已经出现了明显的僵化和刻意的愚弄。
想要真正意义上的，具备完整现代性的社会，就必须要反美。

[りん ദ്ദി˶ᵒ ᵕ ˂) @Rin350786]

@StockMKTNewz …傻逼