开启时会模糊预览图，关闭后正常显示
LMArena 的机制很简单：用户面对两段匿名模型输出，只凭观感、流畅度、完成度做出选择。 某种程度上比常规 benchmark 更贴近真实使用场景。 这次文心的考试结果，在三个方面得分很高：创意写作第一，复杂长问题理解稳定领先，指令遵循能力：进入一线梯队。这三个维度，恰好组成了一个 agent-ready 语言模型最核心的三角形结构。 特别是创意写作这个点，这其实是最不容易欺骗的维度，它考的是语言建模是否真的贴近人类思维节奏，能不能写出不油腻，有节奏感的段落。 这次我看到的文心不一样的点就在能用中文表达出轻盈克制，并且不空洞的内容。这种能力其实有点稀缺， 更关键的是这还只是 Preview，正式版本会下周百度世界大会上亮相。现在看到的能力，还只是可控泄露版本。
从首次亮相到现在，文心其实经历了几个关键拐点。 每一代的升级，都没有大规模的吹嘘参数量、排行榜，默默的在 B 端市场让语言模型成为一个具备判断力的系统入口。 这次上榜更多像是在验证一个阶段性假设，谁能以人的方式组织任务。 文心大模型5.0-Preview 在创作感/执行精度/理解复杂任务结构这几个点上出现了它自己的策略感。 我从这次榜单看到的是一个国产模型，在不强调民族主义叙事、不靠宣传强度的前提下，站上了对比场的上位区间。 语言的本质是世界观的映射。 大模型的能力边界，其实是对话世界的方式边界。 下周百度世界，再看正式版落地的系统能力表现，才是真正的对话开始。
also relevant: crystallize to proselytize
traveler btw worlds. bias for makers, I heart art + tech! capitalist. EIC a16zcrypto; Editor in Chief a16z + podcast showrunner 2014-2022; fmr WIRED, Xerox PARC
LMArena 排名更新，文心大模型5.0-Preview走到前台 LMArena这么火，但过去我们已经习惯看中文大模型的“追赶节奏”，这件事情其实一直被人诟病。 这次LMArena排行榜更新，反而值得停下来拆拆这个局面。国产模型从定向训练榜单的补课者，开始大规模正面进入竞技场。 百度最新的ERNIE-5.0-Preview-1022 在 LMArena Text 排行榜上全球并列第二、国内最佳，超过 GPT-5-High、Claude-Opus-4.1、GPT-4o 等多个顶级模型。 跟以往的秀参数不同，这次是在盲评投票机制下的对抗。
LMArena 的机制很简单：用户面对两段匿名模型输出，只凭观感、流畅度、完成度做出选择。 某种程度上比常规 benchmark 更贴近真实使用场景。 这次文心的考试结果，在三个方面得分很高：创意写作第一，复杂长问题理解稳定领先，指令遵循能力：进入一线梯队。这三个维度，恰好组成了一个 agent-ready 语言模型最核心的三角形结构。 特别是创意写作这个点，这其实是最不容易欺骗的维度，它考的是语言建模是否真的贴近人类思维节奏，能不能写出不油腻，有节奏感的段落。 这次我看到的文心不一样的点就在能用中文表达出轻盈克制，并且不空洞的内容。这种能力其实有点稀缺， 更关键的是这还只是 Preview，正式版本会下周百度世界大会上亮相。现在看到的能力，还只是可控泄露版本。
this one hit
curious guy creating things @ https://t.co/HXWladhJaA - up and coming wife guy
Jeff Dean 的传奇笑话梗，很多我也看不懂，可能是程序员才懂的幽默 Jeff Dean proved that P=NP when he solved all NP problems in polynomial time on a whiteboard. 杰夫·迪安在白板上用多项式时间解决了所有 NP 问题，证明了 P=NP。 Jeff Dean's PIN is the last 4 digits of pi. 杰夫·迪安的 PIN 是π的最后四位数。 When Jeff gives a seminar at Stanford, it's so crowded Don Knuth has to sit on the floor. (TRUE) 当杰夫在斯坦福大学做讲座时，场面如此拥挤，以至于唐纳德·克努特不得不坐在地上。（真） Jeff Dean once bit a spider, the spider got super powers and C++ readability. 杰夫·迪恩曾经咬过一只蜘蛛，蜘蛛获得了超能力和 C++的可读性。 Once, in early 2002, when the index servers went down, Jeff Dean answered user queries manually for two hours. Evals showed a quality improvement of 5 points. 2002 年初，当索引服务器宕机时，杰夫·迪恩手动回答用户查询了两个小时。评估显示质量提升了 5 分。 Jeff Dean got promoted to level 11 in a system where max level is 10. (TRUE) 杰夫·迪恩在一个最高级别为 10 的系统中晋升到了 11 级。（真） Google Search was Jeff Dean's Noogler Project. 谷歌搜索是杰夫·迪恩的 Noogler 项目。 Jeff Dean has punch card readability. Jeff Dean 具有打卡式的可读性。 Jeff Dean puts his pants on one leg at a time, but if he had more than two legs, you would see that his approach is actually O(log n). Jeff Dean 是一次一条腿地穿裤子，但如果他有超过两条腿，你会看到他的方法实际上是 O(log n)。 Jeff Dean acquired Sawzall readability after writing 58 lines of Sawzall code. As part of his readability review, he pointed out a flaw in the style guide which was promptly corrected by the reviewer. Jeff Dean 在编写了 58 行 Sawzall 代码后获得了 Sawzall 可读性。作为他可读性审查的一部分，他指出风格指南中的一个缺陷，该缺陷被审查者迅速纠正。 Sanjay once asked Jeff Dean if he could keep the entire web in his memory. Due to the noise from his keyboard cooling fan, Jeff Dean misheard slightly and wrote Mustang instead of simply answering "Yes". Sanjay 曾经问 Jeff Dean 是否能将整个网络都记在他的记忆中。由于键盘冷却风扇的噪音，Jeff Dean 稍微听错了，写下了 Mustang 而不是简单地回答"是"。 Jeff Dean compiles and runs his code before submitting, but only to check for compiler and CPU bugs. Jeff Dean 在提交前编译和运行他的代码，但只是为了检查编译器和 CPU 错误。 Unsatisfied with constant time, Jeff Dean created the world's first O(1/n) algorithm. 对常数时间不满意，Jeff Dean 创建了世界上第一个 O(1/n) 算法。 Jeff Dean has binary readability. Jeff Dean 有二进制可读性。 Jeff Dean has binary writability. Jeff Dean 有二进制可写性。 When Jeff Dean goes on vacation, production services across Google mysteriously stop working within a few days. This is actually true. 当杰夫·迪恩去度假时，谷歌的各个生产服务会在几天内神秘地停止工作。这其实是真的。 Jeff Dean once shifted a bit so hard it ended up on another computer. 杰夫·迪恩曾经移动得有点厉害，结果 ended up on 另一台电脑上。 During his own Google interview, Jeff Dean was asked the implications if P=NP were true. He said "P = 0 or N = 1." Then, before the interviewer had even finished laughing, Jeff examined Google's public certificate and wrote the private key on the whiteboard. 在他自己的谷歌面试中，杰夫·迪恩被问及如果 P=NP 为真会有什么影响。他说 "P = 0 或 N = 1。" 然后，在面试官还没笑完之前，杰夫检查了谷歌的公共证书，并在白板上写下了私钥。 You use 10% of your brain. The other 90% is running one of Jeff's mapreduce jobs. 你只使用了你大脑的 10%。剩下的 90% 正在运行杰夫的一个 mapreduce 任务。 Jeff Dean's resume lists the things he hasn't done; it's shorter that way. 杰夫·迪恩的简历列出了他没做过的事情；这样写更简洁。 To Jeff Dean, "NP" means "No Problemo". 对杰夫·迪恩来说，"NP"意味着"没问题"。 Jeff Dean wrote an O(n^2) algorithm once. It was for the Traveling Salesman Problem. 杰夫·迪恩曾经写过一个 O(n²) 算法。它是用于旅行商问题的。 You don't explain your work to Jeff Dean. Jeff Dean explains your work to you. 你不需要向杰夫·迪恩解释你的工作。是杰夫·迪恩向你解释他的工作。 Jeff Dean's resume has so many accomplishments, it has a table of contents. 杰夫·迪恩的简历成就如此之多，甚至有目录。 Jeff Dean was forced to invent asynchronous APIs one day when he optimized a function so that it returned before it was invoked. 有一天，杰夫·迪恩被迫发明了异步 API，因为他优化了一个函数，使其在被调用之前就返回了。 The rate at which Jeff Dean produces code jumped by a factor of 40 in late 2000 when he upgraded his keyboard to USB2.0. 2000 年末，当杰夫·迪恩将键盘升级到 USB2.0 时，他编写代码的速率提高了 40 倍。 When Jeff Dean designs software, he first codes the binary and then writes the source as documentation. 当杰夫·迪恩设计软件时，他会先编写二进制代码，然后将源代码作为文档来编写。 Jeff Dean's Peer Review is what got Larry promoted to CEO. 杰夫·迪恩的同行评审促成了拉里晋升为 CEO。 When God said: "Let there be light!", Jeff Dean was there to do the code review. 当上帝说："让光存在吧！"时，杰夫·迪恩正在做代码评审。 When Graham Bell invented the telephone, he saw a missed call from Jeff Dean. 当格雷厄姆·贝尔发明电话时，他收到了来自杰夫·迪恩的一个未接来电。 Compilers don't warn Jeff Dean. Jeff Dean warns compilers. 编译器不会警告杰夫·迪恩。杰夫·迪恩会警告编译器。 Jeff Dean doesn't exist, he's actually an advanced AI created by Jeff Dean. 杰夫·迪恩不存在，他实际上是杰夫·迪恩创造的高级人工智能。 Jeff Dean's IDE doesn't do code analysis, it does code appreciation. 杰夫·迪恩的 IDE 不做代码分析，它做代码欣赏。 Jeff Dean doesn't use ECC memory. He anticipates cosmic rays and uses them to improve performance. 杰夫·迪恩不用 ECC 内存。他预见到宇宙射线，并利用它们来提升性能。 Jeff Dean once failed a Turing test when he correctly identified the 203rd Fibonacci number in less than a second. 杰夫·迪恩曾经在一次图灵测试中失败，因为他不到一秒就正确识别出了第 203 个斐波那契数。 Jeff Dean invented Bigtable so that he would have a place to send his weekly snippets. 杰夫·迪恩发明了 Bigtable，以便他有一个地方可以发送他的每周片段。 On the zeroth day, Jeff Dean created God. 在零日，杰夫·迪恩创造了上帝。 Jeff Dean once implemented a web server in a single printf() call. Other engineers added thousands of lines of explanatory comments but still don't understand exactly how it works. Today that program is known as Google Web Server. 杰夫·迪恩曾经在一个 printf() 调用中实现了一个 Web 服务器。其他工程师添加了数千行解释性注释，但仍然不完全理解它的工作原理。今天那个程序被称为 Google Web Server。 When Jeff has an ergonomic evaluation, it is for the protection of his keyboard. 当杰夫进行人体工学评估时，是为了保护他的键盘。 Jeff Dean can beat you at connect four. In three moves. Jeff Dean 可以在四子棋中赢过你。在三步之内。 Jeff Dean invented BigTable because his resume was too big to fit anywhere else. Jeff Dean 发明了 BigTable，因为他的简历太大，放不下任何地方。 Jeff Dean took the bite out of Apple's logo. Jeff Dean 咬掉了 Apple 标志的一角。 Chuck Norris can kill you. Jeff Dean can kill -9 you. Chuck Norris 可以杀了你。Jeff Dean 可以 kill -9 你。 Jeff Dean can parse HTML with a regular expression...correctly. Jeff Dean 可以用正则表达式解析 HTML...并且正确地解析。 When Jeff has trouble sleeping, he MapReduces sheep. 当 Jeff 失眠时，他会 MapReduces 羊。 When Jeff Dean fires up the profiler, loops unroll themselves in fear. 当 Jeff Dean 启动分析器时，循环会因恐惧而展开。 When your code has undefined behavior, you get a seg fault and corrupted data. When Jeff Dean's code has undefined behavior, a unicorn rides in on a rainbow and gives everybody free ice cream. 当你的代码有未定义行为时，你会遇到段错误和损坏的数据。当 Jeff Dean 的代码有未定义行为时，一匹独角兽会骑着彩虹而来，给每个人分发免费的冰淇淋。 Jeff doesn't sleep, he just sends SIGSUSPEND to the universe. Jeff 不睡觉，他只是向宇宙发送 SIGSUSPEND 信号。 Jeff got Java readability with only 8 lines of code. Jeff 用仅 8 行代码就实现了 Java 的可读性。 Jeff Dean can instantiate abstract classes. Jeff Dean 可以实例化抽象类。 gcc -O4 sends your code to Jeff Dean for a complete rewrite. gcc -O4 将你的代码发送给 Jeff Dean 进行彻底重写。 Jeff can recite 20,000 digits of pi in 5 hours. He doesn't remember them; he just recomputes them on the fly using only O(log n) space. 杰夫能在 5 小时内背诵圆周率的 20,000 位数字。他并不记住它们；他只是在实时中仅使用 O(log n) 空间重新计算它们。 Jeff Dean remembers only one password. For each site, he concatenates it with the site name, computes its SHA-256 hash, and then types the result. 杰夫只记得一个密码。对于每个网站，他将密码与网站名称连接起来，计算其 SHA-256 哈希值，然后输入结果。 Jeff Dean is still waiting for mathematicians to discover the joke he hid in the digits of pi. 杰夫·迪恩仍在等待数学家发现他藏在圆周率数字中的笑话。 There is no Ctrl key on Jeff Dean's keyboard. Jeff Dean is always in control. Jeff Dean 的键盘上没有 Ctrl 键。Jeff Dean 总是掌控一切。 Jeff Dean was born on December 31, 1969 at 11:48 PM. It took him twelve minutes to implement his first time counter. 杰夫·迪恩于 1969 年 12 月 31 日晚上 11:48 出生。他花了十二分钟来实现他的第一个计时器。 When Jeff Dean says "Hello World", the world says "Hello Jeff". 当 Jeff Dean 说 "Hello World" 时，世界回应 "Hello Jeff"。 Jeff Dean can get 1s out of /dev/zero. Jeff Dean 能从 /dev/zero 中获得 1s。 Jeff Dean simply walks into Mordor. Jeff Dean 轻易走进魔多。 Jeff Dean spent some 20% time on an AI project. That produced Urs Hoelzle. Jeff Dean 在一个 AI 项目上花费了大约 20% 的时间。这造就了 Urs Hoelzle。 Google once had to move out of a datacenter after Jeff Dean accidentally compressed the index so densely that a black hole was formed. Google 曾经不得不搬出数据中心，因为 Jeff Dean 碰巧将索引压缩得太密集，以至于形成了一个黑洞。 Jeff starts his programming sessions with cat > /dev/mem. Jeff 开始他的编程会话时使用 cat > /dev/mem。 The speed of light in a vacuum used to be about 35 mph. Then Jeff Dean spent a weekend optimizing physics. 光在真空中的速度曾经大约是 35 英里每小时。然后 Jeff Dean 花了一个周末优化物理。 When Jeff Dean sends you a code review, it's because he thinks there's something in it you should learn. 当 Jeff Dean 给你进行代码审查时，是因为他认为里面有你应该学习的东西。 Jeff Dean does not sleep(), he wait()s. Jeff Dean 不 sleep()，他 wait()s。 Jeff Dean invented MapReduce so he could sort his fan mail. Jeff Dean 发明了 MapReduce，这样他就能整理他的粉丝邮件。 Once Jeff Dean ordered a list, and the list obeyed him. 曾经有一次，Jeff Dean 下达了一个列表，列表就服从了他的命令。 Chuck Norris is Jeff Dean's 20% project. Chuck Norris 是 Jeff Dean 的 20% 项目。 When your code is killed by SIGJEFF, it never runs again. 当你的代码被 SIGJEFF 杀死时，它就再也跑不起来了。 Jeff Dean's calendar goes straight from March 31st to April 2nd; no one fools Jeff Dean. Jeff Dean 的日历直接从 3 月 31 日跳到 4 月 2 日；没有人能骗 Jeff Dean。 Jeff Dean never has the wrong number; you have the wrong phone. Jeff Dean 从不会有错误的号码；你才是那个有错误的电话。 Jeff Dean has exactly two keys on his keyboard: 0 and 1. Jeff Dean 的键盘上只有两个键：0 和 1。 Errors treat Jeff Dean as a warning. 错误将 Jeff Dean 视为警告。 Cricket matches used to take 5 days, until Jeff optimized them. 板球比赛曾经需要 5 天，直到 Jeff 优化了它们。 Jeff Dean's watch displays seconds since January 1st, 1970. He is never late. Jeff Dean 的手表显示自 1970 年 1 月 1 日以来的秒数。他从不迟到。 Jeff's code is so fast the assembly code needs three HALT opcodes to stop it. 杰夫的代码非常快，需要三个 HALT 指令码才能停止它。 Emacs' preferred editor is Jeff Dean. Emacs 的首选编辑器是 Jeff Dean。 Google: it's basically a Jeff Dean's side project. Google：基本上是一个 Jeff Dean 的副业项目。 Jeff Dean has to unoptimize his code so that reviewers believe it was written by a human. Jeff Dean 不得不让他的代码看起来不那么优化，以便审查者相信它是人类编写的。 Websearch is just a large unittest Jeff wrote for his real app. 网络搜索只是 Jeff 为他的真实应用程序编写的一个大型单元测试。 Jeff Dean doesn't need speakers or headphones. He just types cat *.mp3, glances at the screen, and his brain decodes the music in the background while he works. Jeff Dean 不需要音箱或耳机。他只需输入 cat *.mp3，瞥一眼屏幕，然后在工作时，他的大脑就能解码背景音乐。 Jeff Dean has Perl Readability. (TRUE) Jeff Dean 拥有 Perl 可读性。（真） Jeff Dean quicksorts his laundry. Jeff Dean 用快速排序法来洗衣服。 The OR ELSE construct had to be removed from ISO C after Jeff Dean used it in Mustang and kernels started panicking in terror. 在 Jeff Dean 在 Mustang 中使用 OR ELSE 结构后，该结构不得不从 ISO C 中移除，因为内核开始恐慌地崩溃。 Jeff Dean is not afraid of evil constructors. They are afraid of him. 杰夫·迪恩并不害怕恶意的构造函数。是它们害怕他。 Jeff Dean doesn't write bugs, just features you are unable to understand. Jeff Dean 不写 bug，只是你无法理解的功能。 Jeff Dean eschews both Emacs and VI. He types his code into zcat, because it's faster that way. Jeff Dean 都不使用 Emacs 和 VI。他把代码输入到 zcat 中，因为这样更快。 When Jeff Dean sends an Ethernet frame, there are no collisions because the competing frames retreat back up into the buffer memory on their source network cards. 当 Jeff Dean 发送一个以太网帧时，不会发生冲突，因为竞争的帧会退回到源网络卡的缓冲内存中。 Jeff once simultaneously reduced all binary sizes by 3% and raised the severity of a previously known low-priority Python bug to critical-priority in a single change that contained no Python code. Jeff 曾经在一次修改中同时将所有二进制文件大小减少了 3%，并将一个之前已知的低优先级 Python bug 提升为高优先级，而这次修改中不包含任何 Python 代码。 One day, Jeff Dean grabbed his Etch-a-Sketch instead of his laptop on his way out the door. On his way back home to get his real laptop, he programmed the Etch-a-Sketch to play Tetris. 一天，杰夫·迪安出门时拿了他的画线描板而不是笔记本电脑。在他回家拿真笔记本电脑的路上，他编程让画线描板玩俄罗斯方块。 The x86-64 spec includes several undocumented instructions marked private use. They are actually for Jeff Dean's use. x86-64 规范中包含一些标记为私有的未公开指令。它们实际上是供杰夫·迪安使用的。 Knuth mailed a copy of The Art of Computer Programming to Google. Jeff Dean autographed it and mailed it back. 克努特给谷歌寄了一本《计算机程序设计艺术》。杰夫·迪安在上面签名后寄了回去。 When he heard that Jeff Dean's autobiography would be exclusive to the platform, Richard Stallman bought a Kindle. 当听说杰夫·迪恩的自传将独家在平台上发布时，理查德·斯托曼购买了一台 Kindle。 Jeff Dean can losslessly compress random data. 杰夫·迪恩可以无损压缩随机数据。 When asked if the facts about him are true, Jeff Dean responded with "111111". While the interviewer was still trying to figure out what he meant, he clarified, "every single bit of it is true." 当被问及关于他的那些事实是否属实时，杰夫·迪恩回答说"111111"。而记者还在试图弄清楚他的意思时，他澄清说"每一个字都是真的"。 Jeff Dean mines bitcoins. In his head. 杰夫·迪恩在脑海里挖比特币。
喜欢摇滚乐、爱钓鱼的PM 网站：https://t.co/vnUpLt752o
the disappointing part is i dont think theres anything to actually reliably short (aside from like, attention spans and the birth rate and gen alpha and so on) so i feel kinda stupid loving this movie so much throughout the ai bubble
the equivalent of the bailout in this case is technocapital realizing it doesnt actually need the next generation of founders. its okay to cook gen alpha until they cant even form a coherent sentence, because you wont need them by the time theyre 30 to continue economic growth