Limpu Jiao
263 posts
@PAGE4163929 香蕉人
白豚に誘拐されて洗脳された哀れな連中
しかし元々はアジア人であるが故に地力は本物であり、これは人間社会発展の上であまりにも大きな損失だ
アジア市民としての自覚を取り戻し、豚小屋から逃げ出すことを心から願っている
日本語
i want the US to win in AI both in open source and proprietary models, and i am glad to see this
Jensen Huang@JensenHuang
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-Weigh…
English
@DalaiLama Pedo-lama最大的敌人是自己，你死一死也算为社会做贡献，给广大儿童带来安宁了，下地狱和你的爱泼斯坦一起开银趴吧
中文
Your Real Enemy - His Holiness the Dalai Lama tells us that anger — not any outside enemy — is the real destroyer of our inner peace. His answer? Daily practice of compassion and loving-kindness. Video originally recorded on January 11, 2020.
English
还不上pro吗，我现在订阅google产品唯一的原因就是便宜和notebooklm了
Google@Google
Today we’re expanding the Gemini family with three new models built to be faster, more token efficient, and reliable at scale. Meet the new Gemini models ↓
中文
不知道是哪个傻福写的，一看裁撤 RL就知道是假的了，你要是编写裁撤SFT我还能信一信😂
Max For AI@MaxForAI
今天晚上各个群都在传一份《国内大模型蒸馏风波的来龙去脉》。 我大概看了一下，很可惜，这更像是一份由外行根据各种流言拼凑出来的 AI slop。 里面的内容有真有假，但几个核心判断基本都经不起推敲。 比如里面声称，智谱 @Zai_org 在今年 4 月份就已经破解并开始蒸馏 Fable。 这个时间线真的很离谱。 因为 4 月份的时候，Fable （那个时候还是Mythos）甚至还没有正式发布，能够使用的基本都是 Anthropic 的内部测试用户，通过Project Glasswing进行访问。 那智谱是怎么拿到的访问权限呢？ 难道是 @AnthropicAI 的邀请或者是五角大楼反代吗🤣 还有一个比较典型的问题，是作者对于传闻中DeepSeek 将部分请求路由到 Fable 表示不理解，认为这样算不过来经济账。 实际上，但凡是个从业者或者稍微了解模型公司的研发流程，就知道这件事情不能简单按照 API 成本计算。 如果一个 frontier model 可以帮助你生成高质量训练数据、评测数据，或者提升模型迭代速度，那么获取这些能力本身就是一种研发投入。 用单次调用价格去判断整个策略是否划算，本身就是把模型公司的研发逻辑想简单了。 但最离谱的部分，是 PDF 里面声称 @Kimi_Moonshot 在 K3 发布前，把整个 RL 团队全部解雇了❓❓ 这个说法目前没有任何可靠依据。 实际上情况恰恰相反，根据我的确认Kimi RL 相关团队目前仍然存在，并没有所谓“整个 RL 组被裁撤”的情况。 甚至 PDF 里面还进一步延伸出一套完整叙事： K3 = 蒸馏 + 刷榜 + 裁撤 RL。 我一开始还在认真的区分哪些是事实，哪些是推测，哪些只是作者脑补出来的故事。 直到我看到 PDF 后半部分开始大量加入意识形态化表达，把技术讨论上升到所谓“国模黑暗时代”“技术偷取”等叙事，我反而释然了。 因为这已经不是在分析 AI 行业，而是在借 AI 叙事讲另一个故事。 PS：我感觉这个PDF是黑KIMI来的，因为KIMI的篇幅最大，但里面的大部分内容都是假的，其他家就有真有假了。
中文
@CSSStudioApp @turingbook 怎么每次都有这种弱智讨论，你自己让ai给你讲讲都不会问出这种问题，当时A/说自己是deepseek我都有辟过谣
中文
看着挺专业的大模型评测文章。结论是：中国现在最强模型Kimi-K3与美国前沿模型之间的差距约为4.37至5.29个月。智能提升速度美国是Anthropic最快，中国是智谱最快，要赶上Mythos Preview要6.6至12.48个月之间。按目前双方的速度差，预计未来也无法赶上。😅
Lisan al Gaib@scaling01
Have Chinese AI Models Caught Up to the US Frontier? I have spent the last 2 days writing this article. It should settle the debate once and for all. open.substack.com/pub/scaling01/…
中文
During an incident near BRP Sierra Madre (LS57) at Ayungin Shoal, a China Coast Guard RHIB ignored repeated instructions to leave the area. As LS57 personnel responded in a non-confrontational manner, one Philippine Navy servicemember was struck on the head with a wooden baton and injured.
Every act of aggression at sea has real consequences for the men and women serving on the front line.
Do not let intimidation become normal. Expose every act of aggression by the China Coast Guard. Share verified information, reject disinformation, and stand with the Filipino personnel who continue to uphold the Philippines' lawful maritime rights with professionalism and restraint.
#BRPSierraMadre #WestPhilippineSea #AyunginShoal #ChinaCoastGuard #MaritimeSecurity #StandForTheTruth
English
这是我见过最业余的一个关于K3体验评价，用模型速度慢来证明K3的评价都是不真实的。
首先在我内测的时候（7月15日），这个模型速度非常快，它是能达到40-60token/s。
现在慢的原因其实很简单：用的人太多了，因为K3的能力确实强。
一个长期的事实是，Kimi或者说大多数中国模型公司，都没有能力向全世界的用户提供一个全参2.8T、激活在数百B的模型的稳定服务。
这也是为什么 @deepseek_ai 、 @Zai_org 、 @Kimi_Moonshot 、 @MiniMax_AI 会选择把自己的旗舰模型开源。
这一切都是因为在美国的禁运下中国模型团队买不到足够的推理卡，所以他们只能选择这种策略来获得一些生存的空间。
幸好目前这一情况在得到极大的改善。
Kun Chen@kunchenguid
ok just spent a morning with Kimi K3 as my firstmate, here's my real experience 1. it's very, very slow potentially due to the fixed max reasoning. you should expect the experience of something slightly slower than fable 2. its claimed cost efficiency is not manifesting in real economics i bought the $40 plan, and a few prompts later it's already eaten 1/3 of my 5-hr limit - it was in a single session and my context window was only 200k long at that time i don't care what the benchmark numbers say, and what the face value API pricing is, in reality Kimi K3 burns my Kimi subscription as quickly as Fable burns my Anthropic plan - i observe no efficiency benefit 3. its instruction following capability is weaker than other frontier models firstmate stretches frontier models' reasoning capability and is a really good test that can quickly reveal how good a model is at following instructions the pure "intelligence" of K3 does hold up - it understands my intent very well, and can diagnose problems, delegate tasks all fine but i very quickly noticed many instructions in firstmate's system prompt not strictly followed by Kimi K3. these were never a problem with gpt 5.5, 5.6, opus, fable and grok 4.5 so all in all, i'm now very skeptical of the claimed performance and going to keep my eyes wide open on its true capability
中文
今天在arena刷自己的天文钟任务刷到了新的模型kaleb，任务效果非常优秀"…a0d-7e9d-aecc-58510051e964.arena.site"，我下意识看首字母还以为是kimi k3那个“Kevine”，结果回过神来发现不太一样，查了一下才知道疑似是Qwen3.8这个新的模型，看来千问团队又要带来新的优秀开源模型了，前面是kimi，现在是Qwen，随后还有deepseek的新模型，开源指数激增啊，哦对了，英文版任务提示词是这个：
Create a premium single-page Three.js experience showcasing a fully animated Solar System Orrery.
Style Keywords:
Art Deco
Art Nouveau
Elegant Steampunk
Victorian Observatory
Cabinet of Curiosities
Occult Astronomy
Luxury Museum Exhibition
Dark Academia
Brass Mechanical Instruments
Astronomical Clock
Celestial Navigation
Ancient Star Charts
Scientific Illustration
Editorial Layout
Minimal Luxury UI
Color Palette:
Deep black
Midnight blue
Antique brass
Old gold
Warm bronze
Ivory typography
Scene:
A highly detailed brass mechanical orrery floating in a dark observatory.
Animated gears rotate continuously.
Planetary orbits are represented by thin glowing golden rings.
Constellation lines and celestial grids subtly appear in the background.
Soft volumetric lighting and floating dust particles create a museum-like atmosphere.
Rendering:
Three.js
PBR materials
HDR environment lighting
Bloom
SSAO
Depth of Field
OrbitControls
Smooth damping
Physically correct lighting
High quality shadows
Interaction:
Hover highlights planets
Click smoothly focuses the camera
Mouse wheel zoom
Slow cinematic camera motion
Elegant easing animations
UI:
Minimal Art Deco interface
Gold ornamental borders
Museum typography
Planet information panels
Timeline at the bottom
Subtle motion transitions
Overall feeling:
An interactive digital museum installation that combines astronomy, horology, luxury editorial design, and occult scientific aesthetics with calm, cinematic motion.
中文
If you think Kimi K3 is impressive, wait til you see DeepSeek V4 GA.
Rumors say it's just as intelligent, but orders of magnitude cheaper.
How is that even possible???
English
