xLight (@xlight) profile picture

80初!喜欢玩最前沿的科技玩具,前提是有民用价值。崇尚Simple Amazing的设计。家有医生

角门
Joined June 2007
Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)Photo by xLight (@xlight)
Fuck you, @ArtificialAnlys Your new index says Claude Haiku 5.5 is smarter than DeepSeek V4.1 Flash 1017 and GLM-5.3. Haiku is Anthropic's small cheap model. DeepSeek 1017 is the model Cursor put on its cost frontier this month, and Cursor says no model scores higher at those prices. I run it every day on my 4x 3090. I dont believe for a second that Haiku beats it. Look at how they got there. 60% of the index now runs on private tests nobody outside can see. GDPval-AA, SciCode and the long context test are gone. In comes a private coding benchmark from Cognition. And the open weights gap goes from 3 points to 12 in 1 release, 4x. Released open models are now 26 points behind the top. 1 day before v5, on the old version, they gave Ling 3.1 Flash 1019 a 44. Now the best released open model gets 32. The weights didnt change. The tests did, and you cant see the tests. This is a bought benchmark. A leaderboard where you cant check a single question, that widens the gap 4x overnight and puts Haiku above DeepSeek, is marketing for closed labs. Its not measurement. I hope at least you were paid enough money to publish such garbage. Liars. Publish the questions or stop calling it an index.
Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers. Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale. New to the Intelligence Index: ➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change ➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment Upgraded: ➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions ➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised ➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0 Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1 Key results: ➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50 ➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36) ➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04
207
118
33
2,833
172,638
xLight retweeted
闭关搞出了一个多年前想做的广域网优化协议,正好是七夕,就叫 Queqiao 吧。在 MSRA 的时候直连外网的专线非常爽,一般只能通过香港中转才能达到这种性能。最近测了下发现,中美之间大多数情况下都是 erasure channel + upper bandwidth bound,因此自适应的 FEC 非常好用。 做了一个协议,自己在 Mac 上跑了 3 天非常好用,开源出来。只要在本地和美国的出口网关服务器上分别装上 daemon 就可以用。(Mac/Linux 就是个 daemon 可以搭配现有的客户端使用,iOS/Android 自己开发测试中) 欢迎测试 & star & 提 issue! github.com/bojieli/queqiao
54
65
3
526
55,484
Takaichi's Ambiguity Betrayal of History Eighty-one years ago, Nagasaki was erased in a blinding flash, with 74,000 victims incinerated by a weapon that should have been the last. Three days earlier, Hiroshima had suffered the same fate. Japanese Prime Minister Sanae Takaichi made it clear at the Hiroshima memorial ceremony on Thursday and again at the Nagasaki memorial ceremony on Sunday that the country's neo-militarists have learned nothing from the horrors of the bombings. At Nagasaki's Peace Park, the city's mayor, Shiro Suzuki, called nuclear weapons an "absolute evil". Yet, standing just a few feet away, Takaichi offered only deliberate ambiguity over whether her government will abide by the Three Non-Nuclear Principles. Japan already operates under the protection of the United States' nuclear umbrella. To invoke a purported "external threat" as a rationale for acquiring an independent nuclear deterrent is hardly the "realistic and pragmatic approach" Takaichi describes — rather, it lays bare the unmistakable ambitions of the country's neo-militarist faction. Takaichi will not say she intends to revise the principles, but neither will she promise to preserve them. She speaks only of current policy, deliberately leaving the future open to interpretation. This is the slow-motion dismantling of a commitment that has defined postwar Japan. The hypocrisy is staggering. Here is a prime minister attending the memorials for victims of nuclear hellfire, surrounded by aging hibakusha, victims of atomic destruction — whose average age is over 86 — yet she will not reaffirm the principles these survivors have defended for decades. When the hibakusha protest her stance on "nuclear sharing" with the US, and when a record 128 local assemblies urge her government to uphold the three principles, her response is little more than a verbal shrug. No wonder Moscow called Japan's nuclear memorial act hypocritical — Takaichi equivocates on the non-nuclear principles while refusing to even ask why the bombs fell on Japan. Eighty-one years after Fat Man fell on Nagasaki, Takaichi stood in Peace Park and blurred the edges of a promise generations of Japanese have held sacred. The city's mayor was right: Nuclear deterrence does not eliminate the risk of catastrophe; it institutionalizes it. The Takaichi government is intent on "honoring" the dead by placing their descendants under the shadow of the same weapons that destroyed them. When the last hibakusha is gone, what will remain of their plea? A prime minister who could not bring herself to unequivocally declare: "I promise to uphold the Three Non-Nuclear Principles". That is not realism. It is a betrayal of history. #Hiroshima #PeaceMemorialCeremony #ShinjiroKoizumi #Shinjiro #Koizumi #Nagasaki #hibakusha #ShiroSuzuki #ThreeNonNuclearPrinciples #nuclear #JapaneseMilitarism #Militarism #NeoMilitarism #Japan #TakaichiSanae #Sanae #Takaichi #SanaeTakaichi #ChinaMilitary #ChinaMilBugle #钧正平
24
33
10
193
43,901
xLight retweeted
CHINA drops war film where MYSTERY INVADER WREAKS HAVOC across MIDDLE EAST FOOTAGE ECHOES US’ IRAQ ‘SHOCK and AWE’ CAMPAIGN as FIREBALLS ERUPT and BOMBS PUMMEL CITIES Chinese restaurant STAFF FORCED into SURVIVAL BATTLE MOVIE: ‘Once Upon a Time in the Middle East’
35
340
19
1,816
128,301
外国博主:“日本不是你以为的日本,它藏了70年的另一张脸,今天扒给你看!”
103
125
16
1,341
42,967
日本历史课内部拍摄流出,在日本小学课堂,是这样讲述日本发动战争的原因……
159
40
15
393
126,335
xLight retweeted
大货车右转弯盲区提示灯 建议全球推广
165
153
20
2,571
345,028
Replying to @MJTruthUltra
The most powerful and profitable industry on Earth has a vested interest in death and disease!
18
195
5
1,096
58,295
xLight retweeted
参考 Cube 的SQL 语义编译做一个基于语义建模的 demo github.com/wubx/databend-sem… 如果你对智能问数敢用还存疑虑,那么这个就是你要找的解决方案。 我们可以一起交流一下。
1
2
118
xLight retweeted
SQL是一种万灵药式的通用解决方案。 而聊天记录,更类似一种log
1
111
Teams都比微信加载快,特别是当你放假一段时间不用聊天工具之后,Teams和QQ一样都是等待O(1)时间后就能用,而微信你不登录越久卡的就越久🤪
微信算奇迹的话,那telegram算什么?
7
1
23
10,592
xLight retweeted
好喜欢 sqlite 论战,希望这不是程序员群体最后一次为技术而愤怒
57
10
3
367
74,385
xLight retweeted
Replying to @Phoenixyin13
我有种不祥的预感。 美国突然把最顶级的模型藏起来不给用。。不会是为了保密吧。。 毕竟。。当年中国军工就是这么干的,保密,就是保胜利。。
1
1
1
724
xLight retweeted
#给阿嬷的情书 一部让人哭的喜剧 一部由邻里街坊,奶茶店赞助的电影 一部上映时无人关注的电影 一部有亲情有爱情有友情,唯独没有偷情的电影 一部豆瓣给了9.1 又被新华网推荐的电影 一部可能会超过10亿票房的小制作电影 一生太短,只够爱一个人。 不知何时,中国的文学艺术圈出现了一种状况。 不批判政府,你就不是一个好作家 不揭露阴暗,你就不是一个好记者 不丑化自己的民族,你就不是一个好画家 不得到洋大人的肯定,你就不是一个好导演 实在找不到什么批判的,揭露的,首肯的 那就编出一点东西去批判 那就造出一点东西去揭露 当你说他们胡编的时候,他们梗着脖子说,我这是艺术是创作 当你偃旗息鼓,他们又神秘兮兮的告诉你 创作源于生活 “我写的都是真的哦” 而这部小制作的电影,没有所谓的批判,没有满篇的揭露。更没有丑化矮化自己,估计也没想到洋人爱不爱看。 他是给邻里街坊们看的,他只靠着真诚,就击败了那些师承欧美,沿用好莱坞叙事,善于资本运作,动则几个亿十几个亿的大制作。 也许中国民众恶心了不负责的批评,腻歪了无休止的揭露,厌倦了过时老套的好莱坞叙事。 只想看人与人的真诚 想看那悠悠的岁月,慢慢的思念,单纯与善良。 想看那刻在中国人骨血里的情义 迎风傲雪的侠骨, 取义轻利的仁者, 一诺千金的赤城, 苦守寒窑的坚贞, 卧薪尝胆的隐忍, 精忠报国的热忱。 中国人就不配吃点好的么? 最后用影片里的信中所写作为结尾。 优雅的中文,从来就不需过度的雕琢 行船入夜,恰江上升明月,圆如玉坠,仿若身在故乡, 似与你并肩共赏。 江海万里,心中念你, 便不觉遥远。
61
26
3
131
21,868
xLight retweeted
himem.sys emm386.exe 别招笑了,80后什么没见过
3
1
111
任意应用远程投射技术,CTO把性能又优化了一下,内存立马下降了80% ,多开十多个微信不是问题 你们看视频里的效果。任意的一台电脑,直接把安装在微服里面的克隆微信画面投射到本地来 电商办公需要同时开多个应用的小伙伴,想不想要啊?哈哈哈哈 不仅是微信,还可以开其他应用 不用找替代方案了,这个是懒猫微服独家的😏评论区打1,我来介绍优惠
感谢腾讯赏饭吃,以后微服就叫 “支持云端微信的服务器” 这两天太火爆了, 全部都是因为云端微信这个功能买的,今天最后一车货发了就下班 还在头痛微信把苹果笔记本和苹果手机的存储占满了? 还在头痛多个电脑登录微信相互踢? 来一台我们的微服,只需苹果手机一半的价格,永久解决微信空间占用和多端同步的问题,评论区打1,我来给推友专属优惠!
15
5
1
41
54,176
🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length. 🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models. 🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice. Try it now at chat.deepseek.com via Expert Mode / Instant Mode. API is updated & available today! 📄 Tech Report: huggingface.co/deepseek-ai/D… 🤗 Open Weights: huggingface.co/collections/d… 1/n
1,649
7,433
2,421
45,340
10,172,783
炸裂!DeepSeek 不愧是来搞 AI 平权的 不逊色与 Opus4.6 的能力,极其便宜的价格! 这不是有点东西,这是很有实力啊!🥳
就在我还在疑惑这次 DeepSeek V4 Pro 输出价格怎么这么贵的时候,看到了价格下方的小字..... 所以,不会是真的在等昇腾950 部署到位才推迟这么久发布吧?直到等的等不下去,刚好GPT-5.5发了,就先干一波。等昇腾950超节点到位,再用价格屠杀一波。😅
6
4
1
32
16,896
xLight retweeted
推特是不是更新了首页的算法,感觉变成超级信息茧房了怎么刷都是那几条内容
488
167
48
4,547
791,611
😳我用过最早的tui应该是ucdos/wps😳还有 ccdos/cced
3
1
2
3
3,449