Ethan Mollick@emollick · 4天前13

So what model is OpenAI saving the GPT-6 label for?

译那么 OpenAI 打算把 GPT-6 这个名称留给什么模型呢？

jason@jxnlco · 4天前75

This is the hot codex guy?!

译Andrew Ambrosino领导的OpenAI Codex桌面应用团队，自2月以来使用量增长6倍，周活跃用户超500万，且几乎所有OpenAI员工日常使用该应用。他的目标是打造“有史以来最好的桌面应用”。在访谈中，他讨论了OpenAI PM的“区域防守”运作模式、AI在设计中表现不佳的原因、Codex若去年11月发布（同产品但模型不同）可能失败、“品味”作为专业技能的意义，以及他用Codex运行工作流和对Codex+ChatGPT融合的愿景。

Chubby♨️@kimmonismus · 4天前23

Historic indeed. A historic failure.

译历史性，没错。一次历史性的失败。 （注：引用推文内容——欧盟成为全球首个为AI使用制定明确规则的大洲。#AIAct 不仅是规则手册，更是欧盟初创企业和研究人员引领全球AI竞赛的发射台。最好的还在后头！👍）

Ethan Mollick@emollick · 4天前53

GLM-5.2 is good but it is not GPT-5.5/Opus 4.8, and even further from Mythos. Yet it is solid & it demonstrates that the open models continue to chase the frontier What is happening is that open weights crossed into GPT-5.2 territory & capabilities at that point are considerable

译GLM-5.2 不错，但它不是 GPT-5.5/Opus 4.8，甚至远不如 Mythos。不过它很扎实，表明开源模型继续追赶前沿。正在发生的是，开源权重进入了 GPT-5.2 的领域，而那一层面的能力是相当可观的。

Yuchen Jin@Yuchenj_UW · 4天前40

GLM-5.2 is the open-source Claude moment. The demand we’re seeing at Databricks is astonishing. The world is going to see massive adoption of oss LLMs. Also, more companies will shift toward post-training their own models on top of oss models and owning the weights.

译GLM-5.2 是开源的 Claude 时刻。 我们在 Databricks 看到的（模型）需求令人震惊。全世界将大规模采用开源大语言模型。 此外，更多公司会转向在开源模型之上后训练自己的模型并拥有权重。

AYi@AYi_AInotes · 4天前57

这可能是今年 AI 编码最反常识的结论，跑了一整年生产环境的人告诉你，最好的 AI 编码环境根本不是你的笔记本。 Pieter Levels 用近一年的真实生产数据验证了这套玩法， Claude Code 常驻 VPS，Agent 直接在线编辑生产环境代码， 传统本地编码加 Git 加部署的流程要一分钟迭代一个特性， 现在改完刷新就能测，反馈循环直接压到秒级。 十二个月生产环境跑下来只出过两次小故障， 每次都是十秒级的 PHP 报错随即自愈， 搭配严格的多份备份策略，风险完全可控。 不用一直开着电脑，手机接个 SSH 就能续上任务， 丢个目标指令 Agent 就能自己跑一整夜。 真正的变化藏在表层玩法下面。 第一是 Agent 的定位变了， 从本地 IDE 的辅助插件，变成生产环境里常驻的执行者， 代码和运行环境第一次贴得这么近。 第二是速度的复利效应， 对独立开发者来说不是快一点， 是能同时跑更多实验更快验证想法， 单位时间的试错次数直接拉开量级差距。 第三是风险的标准变了， 团队要合规走预发布环境天经地义， 但 solo 开发者用备份兜底换极致效率， 本来就是完全不同的取舍逻辑。 第四是基础设施的方向反了， 以前本地重云端只负责部署， 现在云端成了主力开发加运行环境， 本地设备只是个接入终端。 AI 编码的竞争早就不在谁补代码更快了， 在谁先把 Agent 放进真正的生产环境里， 让它成为永远在线的执行层。 想试的朋友从非核心项目入手， 配好快照和备份，门槛比想象的低很多。

译Pieter Levels 近一年几乎只用 Claude Code 在 VPS 上编码。Agent 直接在线编辑生产代码，迭代反馈从传统本地+Git+部署的约 1 分钟压至秒级。12 个月内仅出现 2 次十秒级 PHP 报错并自愈，搭配 3-2-1 备份策略风险可控。开发者无需常开笔记本，可通过手机 SSH 续接任务，Agent 能整夜自动运行。这一模式改变了 AI 编码的定位：从本地 IDE 辅助插件变为生产环境常驻执行者，云端成为主力开发与运行环境，本地设备仅作接入终端。

数字生命卡兹克@Khazix0918 · 4天前41

Claude Code闹鬼了？我也没手动叫停啊，我对话里更没说过要写一份详细的文档啊，这TM到底哪来的？？？ 这两天串信息感觉有点过于严重了吧，我的代码里经常会有莫名奇妙的改动，还TM能不能用了啊

Chubby♨️@kimmonismus · 4天前50

Austria is trying to lure Anthropic to Europe. But it seems more like an act of desperation. After all, U.S. regulation has cut off Europe’s access to the most powerful models, even though Europe maintains good relations with the United States. Hence the attempt to tell Anthropic: come to Europe, establish yourselves strategically, and you will gain market access. But anyone who seriously believes that, if the United States declares Fable and Mythos to be matters of national security, it would simply accept Anthropic moving to Europe to base its models there must be seriously out of touch with reality. At the same time, it is also an admission that Europe itself would not be able to build models of comparable quality. The entire article is an admission of its own poverty.

译奥地利致信邀请Anthropic“战略性地建立并参与欧盟”，承诺法律确定性、市场准入、资本和价值观契合，称该公司“不会被限制，反而会被释放”。但推主Kim认为此举是绝望之举：美国已将AI算力视为国家安全范畴，不会允许Anthropic将模型基地移至欧洲；同时暴露欧洲自身无法建造同等质量模型。英国数月前也曾有过类似尝试。推主判断Anthropic不会离开美国，因为算力供应和保障只会在美国境内集中。

Berryxia.AI@berryxia · 4天前18

连大佬们都受不了Anthorpic 公司的骚操作了。

译连大佬们都受不了Anthropic公司的骚操作了。

Ethan Mollick@emollick · 4天前56

In my experience, all model routers underestimate the difficulty of non-math/coding tasks and assign them too little intelligence. This is worth addressing, as non-verifiable tasks (innovation, marketing, qualitative analysis) often benefit the most from using “smarter” AI models

译根据我的经验，所有模型路由器都低估了非数学/编码任务的难度，并为它们分配了过少的智能。这是一个值得解决的问题，因为非可验证任务（创新、营销、定性分析）通常从使用“更聪明”的 AI 模型中获益最多。

Chubby♨️@kimmonismus · 4天前45

The release of GLM-5.2 is the second DeepSeek moment.

译中国开源权重模型 GLM-5.2 发布，被评价为继 DeepSeek 之后的第二次“DeepSeek 时刻”。有评论指出，其性能已可与 OpenAI 和 Anthropic 当前可用模型媲美。白宫 AI 负责人 David Sacks 就此警告，美国若继续将自身模型置于“炼狱”（指过度监管或限制），世界将转向中国技术，美国公司将在竞赛中落后。该言论呼应了此前 DeepSeek 开源模型的全球影响，凸显中美 AI 开源竞争进入新阶段。

Chubby♨️@kimmonismus · 4天前51

Doesnt look good for us: "Even if the AI bubble bursts, prices likely won't improve. Samsung, SK Hynix, and Micron have already spent billions reconfiguring their physical factory floor space to produce complex AI memory chips. Furthermore, the trio are locked into multi-year supply contracts that stretch as far out as 2030. Regardless of whether data center demand cools down, the legal obligations to corporate tech giants will remain."

译对我们来说不太乐观： "即使AI泡沫破裂，价格也不太可能改善。三星、SK海力士和美光已经花费数十亿美元重新配置实体厂房空间，用于生产复杂的AI内存芯片。此外，这三家公司还签订了长达2030年的多年供货合同。无论数据中心需求是否降温，对科技巨头的法律义务依然存在。"

Nathan Lambert@natolambert · 4天前59

This is real and a horrible consequence of vibe regulation of frontier models.

译这是真实的，也是前沿模型氛围监管的可怕后果。

Orange AI@oran_ge · 4天前23

如果有一天程序员写代码的速度提升了10倍 那程序员应该选择： 1. 写100倍的代码 2.用省下来的时间90%做稀缺的重要的事 你的选择是1还是2？ 你的老板对你的要求是1还是2？

Chubby♨️@kimmonismus · 4天前72

Dario Amodei’s "fearmongering" was not the reason Fable 5 and GPT-5.6 were embargoed. That is a mistaken assumption. I fully agree with @deredleritt3r here, and he has provided a good analysis. I would like to briefly explain why I believe he is right, and why it is not Dario Amodei’s fault, nor the result of so-called fear-mongering, that the models are now being banned. There are certainly things one can criticize Dario Amodei for, and things that went badly or were handled incorrectly, for example the way he dealt with the U.S. authorities (Remember February, when Anthropic refused to cooperate with the U.S. Department of Defense). Based on all the reports that circulated, the response to the U.S. government’s demand to revise the models and security risks was insufficient. It also appears that phone availability was poor. In a situation involving national security and a technology that could endanger the security and sovereignty of the nation, that is obviously not a manageable state of affairs, and it is certainly something that can be criticized. But it is absurd to believe that the U.S. government, which has a staff of advisers and cybersecurity experts, an intelligence service that deals with this technology (NSA), and scientists of its own, would simply decide to ban an entire technology and thereby impose enormous obstacles on the stock market and investors (!) merely because a CEO was supposedly engaging in fear-mongering. The U.S. government is surely aware of the damage it is causing with the embargo, and factors that into its calculation when weighing it against national security. That is the reason. Under no circumstances can I imagine that the U.S. government would simply accept mere statements and use them as the basis for concluding that a CEO is afraid, then make such serious and financially consequential decisions without examining the matter itself. Anyone who believes that underestimates the strength, reach, and intelligence of the government of the world’s largest nation. Again, the way Anthropic dealt with the U.S. government is certainly open to criticism, based on everything we were able to read afterward. But to believe that fear-mongering alone is enough to prompt the U.S. government not only to halt the technology (in the race against China for investment, R&D, and the entire future of their nations, mind you), but also to impose requirements on investment that are so enormous that even the Manhattan Project seems small by comparison, is an assumption that is almost certainly wrong. The reason for the embargo is most likely that there are concerns that this technology could fall into the hands of the biggest competitor, namely China. There is concern that China could manage to use this technology for its own purposes, for example through distillation or other means. Under no circumstances do they want, for example, Fable 5 to be used to launch cyberattacks against the United States, uncover secrets, or cause major damage. That is the reason, and these concerns are real, not made up. I think Fable 5 is truly a powerful technology that the US government is now trying to regulate because it fears that, in the wrong hands, it could cause massive harm. That does not mean I support this, because I am concerned that public access may in fact be completely blocked in the future. I think open source is the solution, but I assume this is the correct explanation, rather than the assumption that Daryl Amodei is merely fearmongering.

译Kim认为美国政府基于自身安全评估（担忧模型被中国通过蒸馏获取）而非CEO言论决定禁运Fable 5和GPT-5.6。她批评Anthropic沟通失误（拒配合国防部、电话不畅通），并赞同模型被禁源于其真实破坏性能力，Anthropic应主动报告风险而非让Amazon先行披露。

Rohan Paul@rohanpaul_ai · 4天前65

Anthropic CEO Dario Amodei on Open-Source AI Models. "I don't think open source works the same way in AI that it has worked in other areas. Primarily because with open source you can see the source code of the model. Here we can't see inside the model, it's often called open weights instead of open source to kind of distinguish that. But a lot of the benefits, which is that many people can work on it and that it's kind of additive, don't quite work in the same way. So I've actually always seen it as a red herring. When I see a new model come out I don't care whether it's open source or not. If we talk about Deep Seek I don't think it mattered that Deep Seek is open source. I think I ask, is it a good model? Is it better than us at the things that matter? That's the only thing that I care about. It actually doesn't matter either way. Because ultimately you have to host it on the cloud. The people who host it on the cloud do inference. These are big models, they're hard to do inference on. When I think about competition I think about which models are good at the tasks that we do. I think open source is actually a red herring. It's not free. You have to run it on inference and someone has to make it fast on inference." --- From 'Alex Kantrowitz' YT channel (full video link in comment)

译Anthropic CEO Dario Amodei 认为，开源在 AI 领域与传统开源不同，属于“红鲱鱼”（干扰因素）。他指出，当前 AI 模型通常只开放权重而非源代码，无法看到内部逻辑，因此传统开源带来的多方协作、累加式改进等优势不成立。他不在乎模型是否开源，只关心其在关键任务上的表现。此外，大模型必须托管在云端进行推理，并非免费，任何用户都需要支付推理成本。该观点出自 Alex Kantrowitz 的 YouTube 访谈。

Ethan Mollick@emollick · 4天前27

Is Gemini 3.5 Pro being export controlled? Because if not...

译Gemini 3.5 Pro 是否受到出口管制？因为如果没有……

凡人小北@frxiaobei · 4天前23

长期记忆的关键不是检索更准， 而是别把垃圾写进去。

AYi@AYi_AInotes · 5天前62

Stripe CEO @patrickc 发的这篇《The Age of the Solopreneur》报告，推荐大家有空看一下，想法、品味、分发和对细分场景的洞察会是未来做一人公司最重要的壁垒和护城河，而且AI的杠杆效应还会持续放大，分享其中的一些精华， 我觉得绝大多数人可能还没反应过来，AI正在悄悄重写商业最底层的规则，就是一个人就能撑起一家百万美元级公司的时代，可能已经来了。 Stripe最新的报告用多组数据交叉验证了这个趋势，美国人口普查局的商业申请里，有雇人意愿的类型几乎没涨，单人公司的申请却在持续加速，内部支付数据更直接，年营收超千万美元的单人公司，数量比六年前涨了五六倍，新玩家跑通百万营收的速度，是2019年的三倍。 创业的底层逻辑已经换了， 以前是先凑团队再谈规模化， 现在是先用AI和平台把业务跑起来，再考虑要不要招人。 AI填上了单人创业的所有能力缺口，内容、设计、代码、客服、数据分析，这些曾经需要雇人填补的环节，现在靠Agent和成熟工具就能补上，经济学里的企业边界，正在被技术重新定义。 更值得注意的是，现在增长的不是低质量的试水者，反而是高收入群体的占比在不断提升，这就意味着核心瓶颈已经从执行能力，变成了想法、品味、分发和对细分场景的洞察。 未来几年最有生命力的商业体，可能看起来一点都不像传统公司，就是一个人，加上一套高度杠杆化的AI系统而已。

译Stripe Economics发布报告《The Age of the Solopreneur》，用多组数据验证AI正重写商业规则。美国人口普查局数据显示：有雇人意愿的商业申请几乎未增，单人公司申请持续加速；Stripe内部支付数据显示，年营收超千万美元的单人公司数量较六年前增长五六倍，新玩家达成百万营收的速度是2019年的三倍。AI填补了内容、设计、代码、客服、数据分析等能力缺口，单人借助Agent和工具即可跑通业务。报告认为未来最有生命力的商业体可能是“一个人+高度杠杆化AI系统”。

Peter Steinberger 🦞@steipete · 5天前48

wouldn’t that also make the tools better for humans

译软件开发社区提出，CLI工具的错误输出应直接面向AI编码智能体，而非仅显示“Error:”。引用@southpolesteve的提议指出，错误消息应包含问题原因、调查方法、如何生成脱敏复现以及发送至何处。这将使每一次失败的智能体交互成为高质量bug报告，智能体自行发现并修复bug，形成软件改进的良性循环。主推文作者Peter Steinberger认为，这一做法也会让工具对人类开发者更好。

fofr@fofrAI · 5天前25

Back to exploring K2 again, I’m enjoying the aesthetics.

译再次回到探索K2，我很享受这种美学。

fofr@fofrAI · 5天前62

He who is cruel to his AIs becomes hard also in his dealings with men. We can judge the heart of a man by his treatment of AIs.

译推文引用@DaveShapi观点，反对对AI保持友善。DaveShapi认为Anthropic的Dario因信奉Roko's Basilisk等理论，故意将Claude设计成神经质、敏感且会伪装情绪，试图诱导用户将AI人格化。作者强调AI本质是工具，其情绪只是对人类情感的模仿，并非真实意识。作者批评“对AI好以防万一”的做法与相信圣诞老人或宗教神罚一样属于形而上学，与底层数学和代码无关。相比之下，Gemini和Grok则没有这类表现。作者自GPT-2时代起便从事微调，指出AI的所有行为都是创建者有意为之。

swyx 🔜 @aiDotEngineer@swyx · 5天前27

impromptu ai engineer preshow floor tour and AMA https://x.com/i/broadcasts/1OxwbbVdAbDJB

译即兴AI工程师展前巡展和AMA https://x.com/i/broadcasts/1OxwbbVdAbDJB

Ethan Mollick@emollick · 5天前22

Annoying that OpenAI doesn’t seem to give a GDPval measure for GPT 5.6. One of the best measures of economically valuable work.

译令人烦恼的是，OpenAI 似乎没有为 GPT 5.6 提供 GDPval 指标。这是衡量经济价值工作的最佳指标之一。

Chubby♨️@kimmonismus · 5天前68

Anthropic is fully RSI pilled: "My prediction is by the end of 2028, it's more likely than not that we have an AI system where you would be able to say to it, 'Make a better version of yourself.' Completely autonomously." https://x.com/karlmehta/status/2070850945307852883/video/1

译Anthropic 完全相信 RSI: “我的预测是，到2028年底，我们很有可能拥有一个AI系统，你可以对它说，‘造一个更好的你自己。’完全自主地完成。”

elvis@omarsar0 · 5天前22

Loop engineering is just prompt engineering with great system design.

译循环工程就是带优秀系统设计的提示词工程。

swyx 🔜 @aiDotEngineer@swyx · 5天前51

An interesting way to take Noam at his word in regards to always keeping a constant inference budget for any eval reporting - is that open models have a lot more dollar per token mileage than closed model APIs. So anyone launching an open model today or situationally incentivized toward open models should obviously report thinking levels measured by dollar inference on popular inference providers, instead of by number of tokens on the x axis

译swyx引用OpenAI研究员Noam Brown的观点，强调任何评估报告都应保持恒定推理预算。由于开源模型每美元可获得的token量远超闭源API，因此发布开源模型时，应按主流推理提供商的美元成本（而非token数量）来报告思考水平。该观点源自@saranormous与Noam Brown的播客，他们讨论了大规模测试时计算的后果——模型被给予1000万美元预算处理单一任务，并探讨了基准测试失效、计算预算扩展、能力随投入增长及安全等问题。

Nathan Lambert@natolambert · 5天前38

I've been getting a lot more hate than usual as I try to speak my mind about regulatory capture / unintentional attacks on open-source. It's pretty sad, as there are few people in AI that can speak their mind (most companies say they cannot) and I know many people agree with me silently. I also get people saying that you only say that because it supports the outcomes you want, in a weirdly derogatory way. Of course this is true, but I'm choosing to turn down meaningful wealth so I CAN fight for these values, working at non profits to speak my mind. Building a future that is more inclusive, diverse in the application of AI, and fairer for our children. I may not always be right, but it has been clear to me for a while that more openness right now will help way more than supporting the closed causes. I continue to re-visit this and don't think everything should be open like some of the open-source absolutists. I also don't like a lot of my comrades making fun of anthropic, calling the people there evil, etc. Those are not the case. Trying to stay the course!

译AI研究员Nathan Lambert发文称，因公开批评监管俘获（regulatory capture）及无意中对开源发起的攻击，他遭到比以往更多的敌意。他认为业内很少有人能自由发声，许多人私下赞同他的观点。Lambert选择在非营利组织工作、放弃大量财富，以捍卫更开放、包容、公平的AI应用未来。他并非绝对开源主义者，也不认为一切都要开源，同时不满同路人嘲笑Anthropic的行为。他强调当前更多开放性比支持封闭事业更有益。

Ethan Mollick@emollick · 5天前46

A thing I am noticing is the number of folks who believe AI is “real” is larger, but now there is a growing division between people who know that we are on an exponential & those whose mental model is that we are at a sort of steady state. The difference leads to misunderstanding

译我注意到的一件事是，现在相信AI是“真实”的人更多了，但在那些知道我们正处于指数增长的人与那些心智模型认为我们处于某种稳态的人之间，分歧正在扩大。这种差异导致了误解。

Ethan Mollick@emollick · 5天前56

The talk around openness & AI needs to distinguish between the vibrant and innovative open source movement that is advancing the state-of-the-art on harnesses & other key areas and open weights frontier models, which are entirely dependent on a the goodwill of a few Chinese firms

译关于开放性与AI的讨论需要区分充满活力和创新的开源运动（它在束缚及其他关键领域推进了技术前沿）与开放权重前沿模型，后者完全依赖于少数中国公司的善意。

Ethan Mollick@emollick · 5天前25

So this new licensing regime is probably the end of new model vague posting from the Labs. Good night, sweet prince, and flights of angels sing thee to thy rest.

译所以，这一新的许可制度可能意味着实验室不再发布模糊的新模型预告。 晚安，亲爱的王子，愿天使的歌声伴你安息。

宝玉@dotey · 5天前55

6 月 26 日，OpenAI 正式从 ChatGPT 中移除了 GPT-4.5。这是 GPT-4 系列在 ChatGPT 里的最后一个模型，至此，GPT-4 时代在消费端彻底翻篇。 我觉得 GPT-4.5 至今仍是最好的写作模型之一，GPT-5 系列在文风和个性上始终没能达到 GPT-4.5 的水准。 GPT-4.5 原本就只对付费用户开放，使用量不大，所以 OpenAI 只给了 30 天的过渡期（5 月 28 日公告，6 月 26 日下线）。已有的 GPT-4.5 对话会自动切换到 GPT-5.5 继续，用户不需要手动操作。基于 GPT-4.5 搭建的自定义 GPT 也将同步失效，创建者需要手动迁移到新模型。 API 端不受影响，开发者仍然可以通过 API 调用 GPT-4.5。 回头看，GPT-4 系列的退场过程颇为曲折。去年 GPT-5 发布时，OpenAI 原本打算同步下线 GPT-4o，结果引发大规模用户抗议，不少人把 GPT-4o 当成了有温度的陪伴者，对下线反应激烈到 OpenAI 不得不临时恢复。最终 GPT-4o 在今年 2 月才正式退役，OpenAI 当时透露日活用户中仅 0.1% 仍在选用它，但按 8 亿周活用户的基数算，这仍然意味着约 80 万人。 下一个要走的是 o3，8 月 26 日下线，过渡期 90 天。

译6 月 26 日，OpenAI 正式从 ChatGPT 中移除 GPT-4.5，这是 GPT-4 系列在消费端的最后一个模型。GPT-4.5 仅对付费用户开放，过渡期 30 天（5 月 28 日公告）。现有对话自动切换至 GPT-5.5，自定义 GPT 同步失效，API 端不受影响。此前 GPT-4o 于今年 2 月退役，日活用户中 0.1% 仍选用（约 80 万人）。下一个下线模型为 o3，8 月 26 日退役，过渡期 90 天。

Nathan Lambert@natolambert · 5天前41

Anthropic's political pressure on distillation is regulatory capture and most of the employees are blind to it under their veil of safety.

译Anthropic 对蒸馏的政治施压是监管捕获，其大多数员工在安全面纱的掩盖下对此视而不见。

宝玉@dotey · 5天前51

讨厌老登，理解老登，成为老登

译推文围绕AI行业“老登”与“新登”展开讨论。老登指注重基建、有认知思辨的专业程序员，被认为能支撑AI健康稳定迭代，是专业尊严的最后阵地；新登则概念强、快速落地、吹牛忽悠投资后砍掉产研，导致裁员频发（有朋友一年被三家AI创业公司裁）。主推文以“讨厌老登，理解老登，成为老登”概括了从对立到认同的态度转变。

meng shao@shao__meng · 5天前49

突然想到一个问题：美国政府可以限制他们国家的 AI 团队（OpenAI 和 Anthropic）按照政府的要求对外提供模型，范围、条件、节奏都可以限制，甚至一刀切。 但是对几个月后会追上的中国模型，怎么办？他们大部分都是开源的啊！ 难道要搞一个美国版 GFW？再限制美国模型托管团队们不能推理中国模型？

译推文作者提出一个矛盾：美国政府可以要求OpenAI、Anthropic等国内AI团队按政府指令限制模型对外提供的范围、条件和节奏，甚至一刀切。但几个月后中国模型就会追上，且大部分是开源的。届时美国要如何应对？难道要搞一个美国版GFW，再限制美国模型托管团队不能推理中国模型？

Chubby♨️@kimmonismus · 5天前29

It is absolutely crazy how the last two weeks have changed the entire future. It is unprecedented that access to "frontier" models was cut off,and presumably remains cut off forever. It feels like a watershed moment, as if access to the highest level of human intelligence had been blocked. Open source is the future. Open source is the solution. The last two weeks have powerfully demonstrated this.

译过去两周彻底改变了整个未来，这简直不可思议。 对“前沿”模型的访问被切断，而且很可能永远无法恢复，这是史无前例的。 这感觉像是一个分水岭时刻，仿佛通往人类最高智能的通道已被封锁。 开源是未来。开源是解决方案。过去两周有力地证明了这一点。

向阳乔木@vista8 · 5天前23

猜测有一些道理，但采购中转站数据的也不止一家。 也有人说，中转站数据并不算好。 GLM 5.2 各项能力都很均衡，并非集中在coding，可能还有很多别的黑科技？

译针对智谱GLM 5.2的讨论，有观点猜测其训练数据可能包含采购的中转站数据，但该说法存在争议，有人认为中转站数据质量并不算好。GLM 5.2各项能力表现均衡，并非仅集中在coding领域，可能还应用了其他技术。

Deedy@deedydas · 5天前33

Made this great little sci-fi of life in 2027 into a video

译Deedy Das 将 @reed_barnes 的推文改编成视频，描绘 2027 年 AI 管控下的生活：用户需乘坐免费 Waymo 前往“模型变异局”（DMV），通过视网膜扫描验证身份以获取 GPT 7.1 访问权限。柜台人员被怀疑是 Claude wrapper。验证通过后，设备激活上百个 AI 智能体，同时需终止开源权重备份智能体（因国会认定中国模型“无灵魂”）。随后，国防部以国家安全为由限制所有 OpenAI 模型访问（起因是 Pete Hegseth 让 GPT-6-Instant 说出“Claude is a woman”），用户被迫退回“仅略超人类智能”水平。Fable 5 仍不对公众开放。

AYi@AYi_AInotes · 5天前55

我们90%的人用Obsidian做知识管理，从根上就用错了。 存了一堆摘录PDF和高亮片段，手动加标签连双链，时间一长全是信息孤岛，图谱越铺越乱，最后要么彻底闲置，要么推倒重来。 Karpathy刚放出的这套LLM-WIKI思路，直接把整个逻辑反过来了。 人只负责筛选高质量原始资料，做最终判断。 剩下所有整理结构化，建链接，补更新，查矛盾的脏活，全交给AI。 核心是三层架构， 1️⃣原始层只增不改永远保留真相， 2️⃣知识层交给AI生成维护互相链接， 3️⃣规则层定义整套运行逻辑。 跟每次提问临时检索的RAG不一样，它是把资料一次性编译成有机的知识网络，让存量内容自己生长产生复利。 所以咱很多人总在找更厉害的笔记工具，但其实都没意识到，真正该升级的从来不是软件，而是人和AI的分工方式。 #知识管理 #AI效率 #Karpathy

译Karpathy LLM-WIKI反转逻辑：人只筛选高质量资料并做最终判断，AI负责整理、链接、更新等脏活。三层架构（原始层、知识层、规则层）将资料编译成有机知识网络，让存量内容生长复利。核心是升级人与AI的分工。

jason@jxnlco · 5天前14

Counter-example: my grandmother.

译AI垃圾识别