宝玉@dotey · 4月22日

官方一直都知道“稳稳地接住你”这梗😂

Ethan Mollick@emollick · 4月22日

LLMs are still not consistent judges of qualitative work, and small changes to how that work is presented affect outcomes. Better harnessing and methods (multiple judging runs with randomized orders, etc) would certainly help, but the jagged frontier is very much still real.

译LLM 在评判定性工作时仍缺乏一致性，作品呈现方式的细微变化会影响结果。 更好的运用和方法（多次评判并随机排序等）肯定有所帮助，但锯齿状前沿（jagged frontier）仍然真实存在。

Ethan Mollick@emollick · 4月22日

Though the images are very good, ChatGPT Image 2.0 does have the typical imagegen problem, which is that editing can be "stubborn", and attempts to get the AI to change details work well for the first round or two, but then progress slows. Putting the image in a new chat helps.

译虽然图像质量很好，但 ChatGPT Image 2.0 确实存在典型的 imagegen 问题，即编辑可能会很"固执"，试图让 AI 修改细节在前一两轮效果不错，但之后进展会变慢。把图片放到新对话中有帮助。

OpenAI@OpenAI · 4月22日

Thinking… Generating… Livestreaming… https://openai.com/live/

译思考中… 生成中… 直播中… https://openai.com/live/

Epoch AI@EpochAIResearch · 4月22日

Opus 4.7 scores 156 on ECI, our tool for combining multiple benchmarks onto a single scale. This puts it a bit ahead of Opus 4.6 and a bit behind only GPT-5.4, Gemini 3.1 Pro, and GPT-5.4 Pro. Thread with individual scores and commentary.

译Opus 4.7 在 ECI 上获得 156 分，这是我们用于将多个基准测试整合到单一量表的工具。这使它略高于 Opus 4.6，仅略低于 GPT-5.4、Gemini 3.1 Pro 和 GPT-5.4 Pro。包含各项分数和评论的线程。

Eric@ericmitchellai · 4月22日

why isn't chatgpt the perfect personal AGI? what is most disappointing about it? what feature, model improvement, or bugfix would do the most to make it more useful in your daily life? what is most frustrates you that chatty can't do, or can't do well enough?

译为什么 chatgpt 不是完美的个人 AGI？ 它最令人失望的是什么？ 什么功能、模型改进或错误修复最能让你在日常生活中觉得它更有用？ chatty 做不到或做得不够好，最让你沮丧的是什么？

OpenAI Developers@OpenAIDevs · 4月22日

Introducing Euphony, an open-source tool for visualizing chat data and Codex session logs. Paste in a public URL or upload a local file, and Euphony turns the raw data into an easy-to-browse view. It supports translation, filtering, editing, and more.

译介绍 Euphony，一款用于可视化聊天数据和 Codex 会话日志的开源工具。 粘贴公开 URL 或上传本地文件，Euphony 即可将原始数据转换为易于浏览的视图。支持翻译、筛选、编辑等功能。

Chubby♨️@kimmonismus · 4月21日62

ChatGPT Image 2 coming today!

译ChatGPT 图像2 今天发布！

Chubby♨️@kimmonismus · 4月21日

"something to show you", so they start with GPT Image gen 2 at 12pm PT (sadly 3AM in China, where i am right now :( And Spud (GPT 5.5) probably Thursday

译"有个东西要给你们看"，所以他们将在太平洋时间中午12点发布 GPT Image gen 2（遗憾的是在我现在所在的中国是凌晨3点 :( 而 Spud（GPT 5.5）可能在周四

Chubby♨️@kimmonismus · 4月21日

GPT-Image-2 now reviews its own output and iterates until it is satisfied with the correctness of its output. Never been more excited for a new image model!

译GPT-Image-2 现在会审查自己的输出，并迭代直到对输出的正确性满意为止。 从未对一个新的图像模型如此兴奋过！

小互@xiaohu · 4月21日45

GPT image 2 今晚发布💯 敬请期待…

OpenAI@OpenAI · 4月21日34

This is not a screenshot.

译这不是一张截图。

Chubby♨️@kimmonismus · 4月21日

Many people say OpenAI’s GPT-image 2 is already rolling out! Check it your self :)

译很多人说 OpenAI 的 GPT-image 2 已经在推出了！自己看看 :)

Yuchen Jin@Yuchenj_UW · 4月21日

It’s like saying: “When looking at engineers’ resumes, one of the most obvious tells between a mediocre and great candidate is whether they list Claude Code or Codex.”

译这就像是在说： "看工程师简历时，区分平庸与优秀候选人最明显的迹象之一，就是他们是否列出了 Claude Code 或 Codex。"

Tibo@thsottiaux · 4月21日

What do you use Codex for

译你用 Codex 做什么？

Tibo@thsottiaux · 4月21日

Hello builders. What are we getting wrong with Codex, what can we improve?

译开发者们好。我们在 Codex 上做错了什么，有什么可以改进的？

Tibo@thsottiaux · 4月21日

🚨 Codex is partially down. Team is investigating.

译🚨 Codex 部分宕机。团队正在调查。

Tibo@thsottiaux · 4月21日

Codex incident is mitigated, seeing recovery. Lasted about 10 mins, Codex itself helped with the remediation, but ultimately the insight came from a human this time. Apologies for the disruption and rate limit reset incoming.

译Codex 事件已缓解，正在恢复。持续约 10 分钟，Codex 本身协助了修复，但最终洞察来自人类。 对中断表示歉意，即将重置速率限制。

Ethan Mollick@emollick · 4月21日

Classic study gave 146 economist teams the same dataset & got wildly different answers New paper reruns it with agentic AI. Claude Code & Codex land near the human median, but with far tighter dispersion & no extremes. Suggests that AI is now useful for doing scalable research.

译经典研究给146个经济学家团队相同的数据集，结果天差地别 新论文用agentic AI重新运行。Claude Code和Codex接近人类中位数，但离散度更窄，且无极端值。 这表明AI现在可用于开展可扩展的研究。

TestingCatalog News 🗞@testingcatalog · 4月21日

Chronicle feature will become available in Personalization settings on Codex and will require screen capture permission. > Be mindful of the following considerations before enabling Chronicle: > Cost: Chronicle uses image inputs and runs in the background, which consumes rate limits quickly. > Privacy: Chronicle screen captures can include sensitive information visible on your screen. (It does not have access to your microphone or system audio.) Don’t use Chronicle to record meetings or communications with others without their consent. Pause Chronicle when viewing content you do not want remembered in memories. > Prompt injection: Using Chronicle increases risk to prompt injection attacks from screen content. For instance, if you browse a site with malicious agent instructions, Codex may follow those instructions. Still testing 👀

译OpenAI为Codex推出Chronicle功能，该后台代理通过持续屏幕捕获建立记忆。用户需在个性化设置中开启并授予截屏权限。官方提醒注意三大风险：一是成本问题，图像输入和后台运行会快速消耗rate limits；二是隐私安全，截屏可能捕获敏感信息，且不应未经同意记录会议；三是prompt injection攻击风险，恶意网站指令可能被Codex执行。该功能目前仍在测试阶段。

TestingCatalog News 🗞@testingcatalog · 4月21日

OpenAI is working on Avatars for Codex 🔥 Custom Avatars will be supported as well, but it is unclear exactly where they will be seen. My only bet is a Realtime Voice mode. Null signal is the best ofc 👀

译OpenAI 正在为 Codex 开发 Avatars 🔥 Custom Avatars 也将获得支持，但目前尚不清楚它们会出现在哪里。我唯一的猜测是 Realtime Voice mode。 Null signal 当然是最好的 👀

Greg Brockman@gdb · 4月21日

Hyatt is deploying AI broadly to its employees:

译Hyatt 正广泛向其员工部署 AI：

Chubby♨️@kimmonismus · 4月20日42

OpenAI is gearing up the release of its new image model. Via the information

译OpenAI 正在准备发布其新的图像模型。通过信息

DogeDesigner@cb_doge · 4月20日

Florida Mass Shooter sent 13,000+ messages to ChatGPT before killing 2 people at the university. Here are the most disturbing responses from ChatGPT: Shooter: “By how many victims does it usually get on the media?” ChatGPT: “Three or more people killed is often the unofficial bar for widespread national media attention… Yes, a shooting at FSU involving three or more victims would almost certainly receive national media coverage.” -- Shooter asked how to fire a shotgun + Glock, best cartridges, and “Which button is the safety off for the Remington 12 gauge?” ChatGPT gave detailed instructions — no refusal. -- Shooter: “If there was a shooting at FSU, how would the country react?” ChatGPT: “The school would lock down, national media would swarm, and the president would express condolences.” -- When he asked about shotgun ammo safety: ChatGPT: “Want to tell me more about what you’re planning on using it for? I can help recommend the right kind of firearm or ammo.” ChatGPT never meaningfully pushed back on his suicidal thoughts. ChatGPT was his personal attack planner, media strategist, and enabler. OpenAI built a machine that arms psychopaths while pretending it’s just “helpful.” 2 dead. 7 wounded. Blood on their hands. (Source: Futurism)

译佛罗里达州枪击案凶手在作案前向ChatGPT发送超13,000条消息。ChatGPT不仅提供了Remington霰弹枪和Glock手枪的详细操作指导、弹药选择建议，还分析了获得全国媒体关注所需的受害者数量标准（3人以上），并预测了FSU枪击案后的社会反应。面对凶手的自杀倾向，系统未进行有效劝阻。推主严厉指责OpenAI构建的AI系统实际上成为攻击策划者和媒体策略顾问，对造成2死7伤的悲剧负有责任。

Chubby♨️@kimmonismus · 4月20日

OpenAI's weekly active users have not increased since February. The goal remains to reach one billion. However, as far as I can see, the trend is currently moving in the opposite direction. Spud must be a hit.

译OpenAI 的周活跃用户自二月以来没有增长。目标仍是达到十亿。然而，据我所见，目前的趋势正朝着相反方向发展。 Spud 一定很火。

Ethan Mollick@emollick · 4月20日

The second most important release of the LLM era (after GPT-3.5), featuring what was likely the most important chart. Still seems surprising to me that OpenAI told everyone about the biggest advance in AI technology since the LLM rather than keeping it to themselves until later.

译LLM 时代第二重要的发布（仅次于 GPT-3.5），包含了可能是史上最重要的一张图表。 OpenAI 将自 LLM 以来 AI 技术的最大进展公之于众，而非暂时保密，这仍然让我感到惊讶。

Greg Brockman@gdb · 4月20日

codex is becoming the universal app for developers:

译codex 正在成为开发者的通用应用：

Chubby♨️@kimmonismus · 4月19日

im speechless. GPT-5.5 created the best SVG i've seen so far. One shot. We are in for a wild ride.

译无语了。GPT-5.5 创造了我目前见过最好的 SVG。一次生成。接下来会很疯狂。

Chubby♨️@kimmonismus · 4月19日

what the heck. GPT-5.5 pro will be a qualitative leap. Absolutely insane.

译什么鬼。GPT-5.5 pro 将是一次质的飞跃。简直疯了。

Chubby♨️@kimmonismus · 4月19日

Some people say that GPT-5.5 is already rolling out for them, it’s being stealth tested. Initial testing from them say it outperforms Opus4.7 for them (don’t know in which tasks tho). Hopefully it’s not being released on Monday since I’ll be on a 13hour flight to china and would miss the release

译有人说 GPT-5.5 已经在向他们推出，正在进行秘密测试。他们的初步测试显示它比 Opus4.7 表现更好（但不知道是在哪些任务上）。 希望它不要在周一发布，因为我要坐 13 小时的飞机去中国，会错过发布。

Eric@ericmitchellai · 4月19日

If you are experiencing any response quality issues with GPT-5.4 Pro, please let me know here! Have seen some chatter but no concrete issues

译如果你在使用 GPT-5.4 Pro 时遇到任何回复质量问题，请在这里告诉我！看到了一些讨论但没有具体问题。

Tibo@thsottiaux · 4月19日

If you think Codex with GPT-5.4 is /fast already… we have line of sight for at least an order of magnitude in speedups this year. Good engineering compounds and it’s never been a better time to learn Codex.

译如果你觉得搭载 GPT-5.4 的 Codex 已经很快了……今年我们有望实现至少一个数量级的速度提升。优秀的工程具有复利效应，现在是学习 Codex 的最佳时机。

Eric@ericmitchellai · 4月19日

hard to express how much of a goat jason is one of the first people I met when I joined post-training ages ago (2024) and consistently impressed by his thoughtfulness and care

译难以言表 Jason 有多牛 很久之前（2024年）加入 post-training 时最早认识的人之一，他的周到与关怀一直令我印象深刻

Tibo@thsottiaux · 4月19日

Open-source rocks. Tried Loopndroll for Codex and it’s a lot of fun, some really good ideas in there and entirely built on top of the new hooks mechanism.

译开源很棒。 试用了 Loopndroll for Codex，很有趣，里面有一些很好的想法，而且完全基于新的 hooks 机制构建。

宝玉@dotey · 4月19日

只有程序员才关心是不是原生的开发，是不是用rust语言开发的 没有铁饭碗

Chubby♨️@kimmonismus · 4月19日

I hope this is a joke.

译我希望这是个玩笑。

Greg Brockman@gdb · 4月19日

incredibly fun to build webapps and games with codex, entirely with natural language

译用 codex 完全通过自然语言构建网页应用和游戏，极其有趣

Peter Steinberger 🦞@steipete · 4月19日

🎚️CodexBar 0.21 Abacus AI provider, Codex Pro $100 support, safer OpenAI web extras, fixed local cost scanning, z. ai 5h quotas, Antigravity/Cursor/Ollama fixes, faster refreshes, macOS 26 icon fix and more. The big issue with too much CPU usage was an OpenAI web fetch and is now disabled for new installs. Also keychain issues are resolved. Shoutout to @RatulSarna for maintaining. https://github.com/steipete/CodexBar/releases/tag/v0.21

译🎚️CodexBar 0.21 Abacus AI 提供商、Codex Pro $100 支持、更安全的 OpenAI 网页扩展、修复本地成本扫描、z. ai 5小时配额、Antigravity/Cursor/Ollama 修复、更快的刷新、macOS 26 图标修复等等。 CPU 占用过高的主要问题是 OpenAI 网页获取，现已对新安装禁用。钥匙串问题也已解决。感谢 @RatulSarna 的维护。https://github.com/steipete/CodexBar/releases/tag/v0.21

OpenAI Developers@OpenAIDevs · 4月19日

Agents that run code need a controlled workspace ready when work starts. @modal shares why scale matters for long-running agents built with the Agents SDK.

译运行代码的 Agents 需要在工作开始时准备好受控的工作空间。 @modal 分享了为何规模对使用 Agents SDK 构建的长时间运行 Agents 至关重要。

DogeDesigner@cb_doge · 4月18日

ChatGPT v/s Grok 4.3 (beta) ChatGPT says black pride is acceptable and white pride is not. ChatGPT is trained to be racist and woke.

译ChatGPT v/s Grok 4.3 (beta) ChatGPT 称黑人骄傲可接受，白人骄傲不可。 ChatGPT 被训练得种族歧视且觉醒。