AIHOT

meng shao@shao__meng · 3天前60

OpenAI Codex 是不是挖来了 PDD 增长负责人 😂 OpenAI 推出可保存（bank/save）速率限制重置的功能，用户可将重置额度留存至后续自行使用，而非必须在重置窗口内立即消耗。 Go、Plus、Pro 和 Business 用户立即获得 1 次免费重置可银行储存。 Plus 和 Pro 用户可邀请最多 3 位好友试用 Codex。好友首次发送 Codex 消息时，双方各获得一次额外可银行重置（30 天内可兑现），限时两周 -- 这也太 PDD 了，朋友们帮砍一刀？

宝玉@dotey · 3天前37

收到 ChatGPT pro 的纪念笔了，感谢 @ChatGPTapp ❤️

译OpenAI 为 ChatGPT Pro 早期订阅者送出限量版纪念笔。福利面向首批 4000 名填表申领的 Pro 用户，填表不保证资格，部分国家可能无法配送。主推文作者@dotey 已收到并致谢。

jason@jxnlco · 3天前75

Great ship by the growth team! Give @abraibrai a follow!

译OpenAI 宣布，用户现可保存 Codex 速率限制重置以供之后使用。Go、Plus、Pro 及 Business 用户将获得一次免费重置。

jason@jxnlco · 3天前66

Great ship by the browser team! Make sure to give @JamesZmSun a follow!

译浏览器团队干得漂亮！记得关注 @JamesZmSun！（摘要：OpenAI 为 Chrome 和 Codex 应用内浏览器推出开发者模式，Codex 可通过 Chrome DevTools 协议分析 JavaScript 性能并检查控制台输出、网络流量和页面状态。）

OpenAI Developers@OpenAIDevs · 3天前53

Invite a friend to Codex and add another reset to the bank. When they send their first Codex message, you’ll both bank one to use when you need it. Rate limit banking is rolling out to Go, Plus, Pro, and Business users, with the first reset on us. https://x.com/OpenAI/status/2065225362544726371

译OpenAI 为 Codex 推出“邀请好友”功能：用户邀请朋友加入 Codex，对方发送第一条消息后，双方各获得一次速率限制（rate limit）重置额度，可存入“银行”留待需要时使用。该功能即日起向 Go、Plus、Pro 和 Business 用户逐步开放，每人首条重置免费。引用推文指出，用户现在可自主保存 rate limit 重置，不再受限于固定时间点。

OpenAI Developers@OpenAIDevs · 3天前69

Introducing developer mode for browser use in Chrome and the Codex in-app browser. Codex can use the Chrome DevTools Protocol (CDP) to debug browser issues by profiling JavaScript performance and inspecting console output, network traffic, and page state.

译为 Chrome 和 Codex 内置浏览器引入开发者模式。 Codex 可以使用 Chrome DevTools 协议（CDP）来调试浏览器问题，通过分析 JavaScript 性能、检查控制台输出、网络流量和页面状态。

Greg Brockman@gdb · 3天前60

For next two weeks, refer your friends to Codex, and you'll bank a rate limit reset:

译接下来两周，推荐朋友使用 Codex，你将获得一次速率限制重置：

Ethan Mollick@emollick · 3天前51

This was a good post though I think it is interesting how much vivid science fiction scenarios have become the new default format for policy papers & financial analysis for AI.

译这是一篇不错的文章，不过我觉得有趣的是，生动的科幻场景已成为AI政策文件与金融分析的新默认格式。

OpenAI@OpenAI · 3天前70

We heard you wanted to use Codex rate limit resets on your own time. Starting today, we’re rolling out the ability to save rate limit resets to use later. We’re starting Go, Plus, Pro, and Business users with one free reset:

译我们听说您希望能在自己方便的时候使用 Codex 速率限制重置。从今天起，我们开始推出将速率限制重置保留到以后使用的功能。我们从 Go、Plus、Pro 和 Business 用户开始，每人提供一次免费重置：

karminski-牙医@karminski3 · 3天前62

另外忘了说了，这个模型支持多模态输入！文本，图片，视频都可以，是真的夯

译Google 发布 Diffusion Gemma，模型大小 26B，激活参数量 4B。与 NVIDIA 合作优化 RTX 4090/5090，5090 每秒可生成 700+ token。支持文本、图片、视频多模态输入。AIME 2026 数学测试达 Gemma4-26B-A4B 的 94%，tau2 bench Agent 测试达 82%。输出质量略逊于传统大模型但速度更快。4bit 量化版本仅需 16GB 显存即可运行。

Rohan Paul@rohanpaul_ai · 3天前67

Dario Amodei's new interview on Bloomberg: The scary part is not when AI does 90% of the job. It is what happens when it learns the last 10%. "We’re already starting to see the beginning of it. There may be some people that it’s not making more productive, and it’s better for the AI to just do the whole thing." And on that topic Claude Code creator Boris Cherny says: "it's very uncomfortable. Artificial intelligence is this force that is far bigger than we are" --- @bbgoriginals From "Bloomberg Originals" YouTube channel, (link in comment)

译Anthropic CEO Dario Amodei 在 Bloomberg 采访中表示，AI 的可怕之处不在于它完成90%的工作，而是学会最后10%的时候。他指出，对于某些人来说，AI 可能不会让他们更高效，不如让 AI 直接做全部。Claude Code 创始人 Boris Cherny 对此评论称，这种感觉非常不舒服，AI 是远比我们强大的力量。

🚨 AI News | TestingCatalog@testingcatalog · 3天前58

GOOGLE 🔥: Gemini Omni Flash will soon be available via APIs for image-to-video, text-to-video, and video editing! Did it get updated again after the release? 👀

译GOOGLE 🔥：Gemini Omni Flash 即将通过 API 提供图像转视频、文本转视频和视频编辑功能！发布后又更新了吗？👀

karminski-牙医@karminski3 · 3天前56

我的使用经验是, one-pass 能力越强(且能在较少的思考下one-pass) 模型才是SOTA的. 要用 agentic coding 才能修复第一次犯的错反而是模型拉夸的表现, 再不济也要在Interleaved thinking过程中修复. agentic coding 是用来解决工程量和运行时问题的. 不是用来修静态检查就行发现的bug的.更简单的说, 你有bug不在thinking中修, 反而非要在n+1次上下文中修复, 是不是骗我买coding plan(x)?

译karminski认为，one-pass能力强（少思考即正确）的模型才是SOTA；需用agentic coding修复首次错误反显模型差，bug应在thinking中修复，而非依赖n+1次上下文，否则有诱导购买coding plan之嫌。@iamai_omni建议测评转向长期任务一致性，可构建loop测评，重点看后续几轮修复表现。

karminski-牙医@karminski3 · 3天前65

单卡 700TPS! Diffusion Gemma 来了! Google 刚刚发布了 Gemma 小模型的 Diffusion 版本! 大小26B, 激活参数量4B, 最重要的是, 这次还跟 NVIDIA 合作针对4090和5090优化了一波, 5090每秒能生成700+token! 给不知道什么是 Diffusion 大模型的同学科普一下, 传统大模型都是一个字一个字吐出来的, 而 Diffusion 大模型则是如同刮奖一样, 是一片一片出来的, 速度高是 Diffusion 大模型的优点. 有得必有失, 缺点当然就是输出质量没有传统大模型好了. 不过这次的 Diffusion Gemma 还是比之前的 Diffusion 文本大模型好不少, AIME 2026(数学能力测试) 能达到 Gemma4-26B-A4B 的94%的水平, 最差的是tau2 bench(考验Agent能力的测试), 也能达到82%. 这个模型大小 4bit 量化版本 16G 显存就能运行了, 另外, 我突发奇想, 这个模型能不能作为 gemma4 dense 模型的草稿模型用来投机解码? 感兴趣的同学可以试试! #diffusiongemma #gemma #gemma4 #google

译Google 推出 Diffusion Gemma，大小 26B、激活参数量 4B，与 NVIDIA 合作针对 RTX 4090/5090 优化，5090 上速度达 700+ token/s。该扩散文本模型以“刮奖式”并行生成而非逐 token 生成，输出质量略逊但优于此前同类模型：AIME 2026（数学）达 Gemma4-26B-A4B 的 94%，tau2 bench（Agent）达 82%。4bit 量化版仅需 16G 显存即可运行。

🚨 AI News | TestingCatalog@testingcatalog · 3天前30

Codex now has an icon selector with new Lite and Dark options available. Classic, Lite, or Dark? 👀

译Codex 现在有一个图标选择器，同时提供了新的 Lite 和 Dark 选项。经典、Lite 还是 Dark？👀

elvis@omarsar0 · 3天前26

My new favorite skill is /learn. I built it to learn any topic at whatever level you like. It combines two of my passions: artifacts and learning. Coming soon to the @dair_ai academy.

译我新最爱的技能是 /learn。我构建了它，用来学习任意话题，任意你喜欢的级别。它结合了我的两个爱好：工件（artifacts）与学习。即将在 @dair_ai 学院推出。

SemiAnalysis@SemiAnalysis_ · 3天前66

Pretraining fundamentally does not make sense anymore for anyone other than frontier labs. Although there are a lot of people at enterprises & startups who have "Pretrainitis" to show “impact” and get promotions, fundamentally, it doesn’t make sense. There is probably higher ROI in partnering with a frontier lab to do prompt engineering, although it isn’t as “sexy” as pretraining.

译预训练从根本上说对前沿实验室以外的任何人都不再有意义。虽然企业和初创公司中有很多人患有"预训练症"以显示"影响力"并获得晋升，但从根本上说，这并不合理。与前沿实验室合作进行提示工程可能会有更高的投资回报率，尽管它不像预训练那样"性感"。

Ethan Mollick@emollick · 3天前61

This is an interesting test, and the frontier models (GPT-5.5 Pro Extended, Claude 5 Fable Max) do fail. They refuse to turn the "three words" into "four" if that fits better Prompting the AI to act like a translator surfaces the problem, but it still avoids changing the wording

译Ethan Mollick 指出，GPT-5.5 Pro Extended 和 Claude 5 Fable Max 在 Beninatto‑Trombetti 翻译测试中失败。该测试要求将“Solo 3 parole: non sei solo”译为英语，同时将 meta‑linguistic 声明从“3 parole”更新为“4 words”（正确译文：“Just 4 words: you are not alone”）。但前沿模型拒绝修改措辞，即使提示扮演翻译角色仍回避变更。Valerio Capraro 认为，Claude 5 Fable 作为最新 LLM 仍无法通过此简单测试，说明 LLM 擅重组已知知识但缺乏真正理解，AGI 仍遥远。

Chubby♨️@kimmonismus · 3天前24

Really curious for Gemini 3.5 Pro. The competition is currently fierce. It needs to be a big release.

译真的很好奇 Gemini 3.5 Pro。当前竞争非常激烈。它需要是一次重大发布。

Chubby♨️@kimmonismus · 3天前13

Anthropic rn

译Anthropic 现在。

Tibo@thsottiaux · 3天前71

Codex 🤟Ona Beyond excited to work with Johannes and team to build the future.

译OpenAI宣布已达成协议收购Ona，其安全的云执行技术将帮助Codex处理更长期的任务（即使笔记本电脑关闭也能运行），并让更多组织安全地部署智能体。收购完成后，Ona将加入OpenAI的Codex团队。Tibo对此表示“无比兴奋，期待与Johannes及团队共同构建未来”。

Orange AI@oran_ge · 3天前39

这是最近我最喜欢的 Fable 5 游戏一次 Prompt 生成的完成度很高非常上头

Replit ⠕@Replit · 3天前68

Replit and @databricks integration just leveled up. Build apps where every user sees only what they should. Your HR analyst can build a full org view for the CEO without ever accessing the underlying data. Public preview is open for sign up! Read more → https://replit.com/blog/databricksjune2026

译Replit 与 @databricks 集成刚刚升级了。构建应用，让每个用户只看到他们应该看到的内容。你的 HR 分析师可以为 CEO 构建完整的组织视图，而无需访问底层数据。公开预览已开放注册！了解更多 → https://replit.com/blog/databricksjune2026

Replit ⠕@Replit · 3天前72

How to prompt like a pro with Replit 🤖 Vague prompts just mean more rewrites. Here's how to get Agent to build the right thing the first time. 🧵 Open thread ↓

译如何像专家一样使用 Replit 进行提示 🤖 模糊的提示词只会导致更多重写。以下是如何让 Agent 第一次就构建出正确的东西。 🧵 展开线程 ↓

elvis@omarsar0 · 3天前25

Got my 10yr old introduced to Codex today. The excitement in his face tells it all. After struggling with Claude Code CLI for a bit, today he was like “this is the future, dad”. The Codex team built a beautiful app.

译今天让我10岁的孩子体验了Codex。他脸上的兴奋说明了一切。在用Claude Code CLI折腾了一会儿之后，今天他说：“这就是未来，爸爸。” Codex团队打造了一个漂亮的应用程序。

elvis@omarsar0 · 3天前74

good. now let's undo the nerf stuff as well

译good. now let's undo the nerf stuff as well （引用推文：Anthropic 在遭受强烈反对后，撤回 Claude Fable 5 秘密降低竞争 AI 研究人员性能的政策。Anthropic 对 WIRED 表示将修改安全措施使其可见，并为此前错误权衡道歉。）

Google Gemini@GeminiApp · 3天前45

Get a closer look at Gemini's new Neural Expressive design language at our next Discord community event. Product Marketing Manager Megan C. will be discussing some of her favorite highlights that help improve the Gemini experience, from dynamic visual responses to seamless mode switching. 👉Join the Discord to watch live: http://discord.gg/gemini 📅 This Friday, June 12 at 11:30 AM PT

译Get a closer look at Gemini's new Neural Expressive design language at our next Discord community event. 在我们的下一次 Discord 社区活动中，近距离了解 Gemini 全新的 Neural Expressive 设计语言。 Product Marketing Manager Megan C. will be discussing some of her favorite highlights that help improve the Gemini experience, from dynamic visual responses to seamless mode switching. 产品营销经理 Megan C. 将讨论她最喜欢的一些亮点，这些亮点有助于改善 Gemini 体验，从动态视觉响应到无缝模式切换。 👉Join the Discord to watch live: http://discord.gg/gemini 👉加入 Discord 观看直播：http://discord.gg/gemini 📅 This Friday, June 12 at 11:30 AM PT 📅 本周五，6月12日，太平洋时间上午11:30

Rohan Paul@rohanpaul_ai · 3天前62

This paper shows an AI improving itself better when it rewrites its setup and updates its model. The problem is that most AI progress still depends on people changing prompts, tools, code, training data, and model weights by hand. The paper’s idea is SIA, a loop where one AI watches how a task agent performs, then either changes the agent’s outer setup or trains the model itself. The outer setup means things like prompts, tools, retry rules, and output parsing, while weight updates mean changing the model’s learned behavior through task feedback. The loop works like this: the task agent tries many answers or programs, the verifier scores them, and those scores become training feedback. Then the system updates a small add-on set of weights called LoRA weights, which changes the model’s behavior without retraining the whole model. So the base model stays mostly the same, but the LoRA adapter learns, “outputs like this got high reward, outputs like that failed.” The authors tested this on 3 very different tasks: Chinese legal charge classification, GPU kernel speed tuning, and single-cell RNA denoising. The combined version beat setup-only improvement on all 3 tasks, reaching 70.1% on LawBench, faster GPU code than the prior best, and 0.289 on denoising. The main lesson is that better scaffolding helps the agent act better, but weight updates help it learn task patterns that prompts and tools alone did not find. ---- Link – arxiv. org/abs/2605.27276 Title: "SIA: Self Improving AI with Harness & Weight Updates"

译该论文提出SIA框架，让AI自动循环改进：一个观察者AI监控任务代理的表现，然后修改其外部设置（提示词、工具、重试规则、输出解析）或通过LoRA权重更新训练模型本身，模型主体不变，仅适配器从任务反馈中学习。在三个任务上测试：中文法律罪名分类（LawBench达70.1%）、GPU内核速度调优（生成代码优于此前最佳）、单细胞RNA降噪（得分0.289）。综合版本在所有任务上超越仅修改设置的方案，表明权重更新能帮助模型学到提示和工具无法发现的模式。

Rohan Paul@rohanpaul_ai · 3天前83

Jeff Bezos on CNBC explains revealed what Prometheus is building. Today his new company Prometheus announced a $12B funding round at a valuation of $41B . Prometheus trying to build an artificial general engineer that can help design and manufacture physical products like engines, medical devices, and electronics. So the target areas are hard physical products like jet engines, chips, bridges, medical devices, consumer electronics, aerospace systems, vehicles, and drug design, where design cycles can take years because every idea has to survive physics, materials, cost, testing, and factory limits. Bezos’ jet-engine example explains it well: asking for the same engine with 10% more thrust can become a 10-year engineering program, and Prometheus wants to shrink that “dream-build” cycle by 10x or more. The $6.2B launch funding gave Prometheus a massive starting base, and the new raise says the company likely needs far more compute, talent, and industrial data before it can prove the product. Their $41B valuation shows that frontier AI is becoming less a software race than a compute procurement race. A company with no broadly shipped product can raise $12 billion at a $41 billion valuation because investors are not only funding a model, they are prepaying for the machines that might make the model possible. The scarce asset is no longer just talent or algorithms, but clustered GPUs, power contracts, cooling, networking, and the operational skill to keep expensive silicon busy. They are proof that demand is arriving faster than infrastructure can be built, and that every frontier funding round quietly turns into a future claim on power, racks, GPUs, and uptime.

译Jeff Bezos 在 CNBC 披露其新公司 Prometheus 的愿景：构建人工通用工程师，设计制造喷气发动机、芯片、医疗设备等硬物理产品，将传统数年设计周期缩短 10 倍以上。公司宣布完成 120 亿美元融资，估值 410 亿美元。初始启动资金 62 亿美元，新一轮融资表明公司需要更多算力、人才和工业数据才能验证产品。410 亿美元估值表明，前沿 AI 已从软件竞赛变为计算采购竞赛——投资者实质在为可能实现模型所需的机器预付费。

Epoch AI@EpochAIResearch · 3天前55

How big a leap is Mythos in cyber capabilities? @timotheechauvin, @AlexBarry4, @js_denain, and @ansonwhho compiled the public evidence and found that while it’s unclear if Mythos was ahead of trend in discovering vulnerabilities, it represents a big jump in exploiting them. 🧵

译Mythos 在网络能力方面有多大的飞跃？ @timotheechauvin、@AlexBarry4、@js_denain 和 @ansonwhho 整理了公开证据，发现虽然尚不清楚 Mythos 在发现漏洞方面是否领先于趋势，但它在利用漏洞方面代表了一次巨大飞跃。🧵

swyx@swyx · 3天前61

congrats to our friends @ona_hq on joining @openai! see their talk here for alpha on what’s next for Codex 👀

译swyx 祝贺 ONA 团队加入 OpenAI，并引用其演讲透露 Codex 的下一步方向。同时引述 @aiDotEngineer 观点：运行 Agent 集群需要三层——Runtime 与编排触发器已解决，但协调层（Agent 间任务交接、验证阶段、继续执行）仍未解决。Stripe 和 RAMP 各自自建了内部方案 Minions 和 Inspect。@loujaybee 指出 GitHub 作为协调层很差——噪音大、仅为人类设计，不适合数百个并行 PR。他提出的修复方案是构建一个 CLI 网关，供本地编码 Agent 调用以检查当前 SDLC 阶段是否已通过、能否继续。

Peter Steinberger 🦞@steipete · 3天前17

Getting Chris to do a PR with Codex!

译让 Chris 用 Codex 做 PR！

elvis@omarsar0 · 3天前59

Same here. Happy with Opus 4.8 (planning) and GPT-5.5 (execution). Also, breaking steps into smaller ones for increasing quality is so underrated. This is why dynamic workflows are a bigger deal than most people think.

译threepointone 使用 Fable 处理一个约 10k 行代码的大 PR，花费 $250，认为不值，更倾向小步骤。Elvis Saravia 认同，表示自己用 Opus 4.8 做规划、GPT-5.5 做执行，并将任务拆解成更小步骤能显著提升质量。他认为动态工作流（dynamic workflows）的重要性被大多数人低估了。其日常仍使用 Cursor AI。

Rohan Paul@rohanpaul_ai · 3天前67

OpenAI is buying Ona to give Codex agents a secure cloud desk that stays open after humans leave. Codex already has 5M weekly users, up 400%, but harder work breaks the old chat pattern because agents need tools, files, credentials, logs, and time. Ona adds persistent cloud workspaces, meaning an agent gets a controlled place to run commands, inspect systems, preserve context, and resume work without depending on one device. The enterprise angle is the real acquisition target: companies want agents inside their own cloud boundary, with scoped credentials, review trails, access limits, and auditable activity. This makes Codexmore like a managed execution layer for tests, bug fixes, refactors, vulnerability work, migrations, and multi-step knowledge tasks.

译OpenAI 宣布收购 Ona，其安全云端执行技术可为 Codex 智能体创建持久云端工作空间——用户离开后，智能体仍可持续运行命令、检查系统、保留上下文并跨设备恢复任务。目前 Codex 周活用户达 500 万（增长 400%）。收购旨在强化企业级部署：智能体可在企业云边界内运行，具备作用域凭证、审核追踪、访问限制和可审计活动，适用于测试、漏洞修复、重构、迁移等多步骤任务。收购完成后，Ona 团队将加入 OpenAI Codex 团队。

Rohan Paul@rohanpaul_ai · 3天前71

Jeff Bezos shuts down AI-induced job loss talk, predicts labor shortage instead Jeff Bezos on CNBC "I think that there’s going to be a labor shortage as a result. Many smart people are saying, oh my God, there are going to be no more radiologists because the AI can read X-rays better than the radiologist can. And there are going to be no more software engineers because the AI can program better than the software engineer can. These people are wrong. What’s really going to happen is that it’s going to elevate all of these people. It’s like, let’s say you’re a software engineer. You’ve been digging out the basement of your house with a shovel, and somebody’s about to hand you a bulldozer. You should be so happy if you’re digging the basement to your house and somebody says, “Hey, how about this? We’re going to have so much productivity in our economy.” ---- From "CNBC Television" YouTube channel, (link in comment)

译杰夫·贝佐斯在 CNBC 反驳“AI 取代人类工作”的观点。他认为，许多人担心 AI 会消灭放射科医生、软件工程师等岗位，但这种看法是错的。AI 实际上会提升这些人的能力，就像挖地下室从铁锹换成推土机一样。他预测结果反而是劳动力短缺，经济生产力将大幅提升。

宝玉@dotey · 3天前73

我用 /goal，长任务很稳定，就不用继续了

译宝玉表示使用 /goal 指令后，长任务运行稳定，不再需要像许多用户那样在AI意外停止时输入“继续”。引用推文指出，不少AI新手不知道AI意外停止时只要发一句“继续”即可恢复任务。宝玉的实践表明，/goal 指令能有效减少此类中断需求。

宝玉@dotey · 3天前62

AI 没有重新定义软件工程，AI 放大了软件工程的重要性

译AI 没有重新定义软件工程，AI 放大了软件工程的重要性 [引用 @arkuy99]：AI 重新定义了软件工程。