@OpenAI:Some of you have been asking very politely. Some of you have just been asking.

GPT-6.2 Astra and Sol are now available.

Super efficient and built to perform where it counts—the real world.

有人很礼貌地问,有人只是在问。

GPT-6.2 Astra 和 Sol 现已可用。

超级高效,专为真实世界表现而建。

[链接:介绍 GPT-6.2 Astra 与 Sol]


@OpenAI:Starting today, GPT-6.2 Astra and Sol are available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.

Pro 500 subscribers can use Ultrafast with either model. For the first time, we’re also bringing Ultrafast to Pro 200 subscribers, starting with GPT-6.2 Sol.

Developers can access gpt-6.2-astra and gpt-6.2-sol in the OpenAI API, with Ultrafast available.

从今天起,GPT-6.2 Astra 和 Sol 在 ChatGPT Work 和 Codex 中对 Plus、Pro、Enterprise 和 Edu 用户开放。

Pro 500 订阅者可用任一模型的 Ultrafast。首次将 Ultrafast 带到 Pro 200 订阅者,从 GPT-6.2 Sol 开始。

开发者可在 OpenAI API 中访问 gpt-6.2-astra 和 gpt-6.2-sol,并支持 Ultrafast。


@OpenAI:GPT-6.2 Astra is our most capable model ever for cyber. It jumps 22 points over GPT-6 Astra on ExploitBench Internal Port.

Due to increased capabilities, we’re launching GPT‑6.2 Astra and Sol with tighter safeguards for higher-risk cybersecurity assistance. We are working to bring these models to Daybreak so approved defenders can access them with reduced refusals.

GPT-6.2 Astra 是我们迄今在网络安全上最强的模型。在 ExploitBench Internal Port 上比 GPT-6 Astra 高出 22 分。

因能力提升,我们以更严格的防护发布 GPT-6.2 Astra 和 Sol 用于高风险网络安全协助。正努力将它们带到 Daybreak,让获批的防御者能以更少拒绝访问。

[图片:相关能力图表]


@thsottiaux:These models are extremely efficient, which you feel in speed to complete tasks, cost and how much further your usage goes

这些模型极其高效,你能在完成任务的速度、成本和用量能撑更远上直接感受到。

[图片:效率相关截图]


@thsottiaux:GPT-6.2 Sol & Astra are finally here. We've also expanded ultrafast to Pro 200 plans for GPT-6.2 Sol, it's my go-to model for most day to day things.

GPT-6.2 Sol 和 Astra 终于来了。我们还把 ultrafast 扩展到 Pro 200 计划用于 GPT-6.2 Sol,它是我日常大多数事情的首选模型。


@ThePrimeagen:most developers would do much better just using one AI chat a time and actually think about what they are doing

too many of y'all larp

大多数开发者如果一次只用一个 AI 聊天并真正思考自己在做什么会好得多。

你们太多人在装。


@ThePrimeagen:people are all trying to figure out what i mean by this

Its quite simple:

There is a bunch of whatever work that goes on (hey move this button here), yeah its easy to make a bunch of small changes

But this is a problem, you never take a step back and deeply think about what you are building and thus we get slop constantly. You don't take the time and figure out what is the best path forward, you just fix the next thing in front of you while having 10 of them going at once.

Take a moment, think about what you are doing.

大家都在试图搞懂我这句话的意思。

其实很简单:

有一堆随便的工作(嘿把这个按钮移到这里),做一堆小改动很容易。

但这是个问题,你从不退一步深入思考你在构建什么,于是不断产出垃圾。你不花时间搞清最佳前进路径,只是同时开着 10 个东西修眼前下一个问题。

停一下,想想你在做什么。


@levelsio:💯

There is no real way to survive for most SaaS companies here because AI is so completely different than any technological advancement before it

Before you could just niche down and specialize on one thing and you'd know for certain the big tech companies would never be able to dedicate enough resources to fill your tiny little niche market of $300M where you could take 1% of the market and make $3M/year

Riches in the niches!

And so you'd avoid going too general, because "you'd spread yourself too thin" and hit "death by a thousand paper cuts"

And general was where big tech companies were great at, they could dedicate huge resources in solving these big problems, like building Google Sheets, or Salesforce (whatever it is they do)

But that was the old days, and it's completely different now

With AI, big tech companies can easily fill any niche, or be general at the same time, they can do the whole range actually, without even spending time on consciously doing that, because they're not offering a particular SaaS offering at all

They're offering you artificial general intelligence (AGI) that can do it all

And as long as they improve their models every time and make them bigger and better, it slowly can do literally anything (okay not yet anything, for anything they'll need physical robots, but they're working on that too of course)

This leaves SaaS companies with a bitter ending:

  • they either do nothing and stay in their niche like they economically should, but that doesn't work anymore now
  • they try go general, and offer EVERYTHING and ANYTHING, and that works for a little bit until customers realize they might as well use whoever makes the models that the SaaS itself uses anyway, which is the big tech and big AI companies, who by the way either offer it free or for much cheaper

Everyone's been checkmated by big tech and big AI

This will be the creative destruction of the century

But there'll be some survivors and then a lot of new companies, but how they'll look nobody knows yet!

大多数 SaaS 公司在这里没有真正的生存之道,因为 AI 与之前任何技术进步完全不同。

以前你可以细分专精一件事,并确信大科技公司永远不会投入足够资源填满你那 3 亿美元的小利基市场,你占 1% 就能年入 300 万美元。

利基里藏着财富!

所以你会避免太通用,因为“会摊太薄”并遭遇“千刀万剕”。

通用是大科技公司擅长的,他们能投入巨大资源解决大问题,比如建 Google Sheets 或 Salesforce。

但那是旧时代,现在完全不同。

有了 AI,大科技公司能轻易填补任何利基,或同时通用,他们实际上能覆盖全范围,甚至不用有意识花时间去做,因为他们根本不提供特定的 SaaS。

他们提供的是能做一切的人工通用智能(AGI)。

只要他们每次改进模型、让它更大更好,它慢慢就能做几乎任何事(还没到任何事,任何事需要物理机器人,但他们当然也在做)。

这让 SaaS 公司面临苦涩结局:

  • 要么什么都不做、像经济上该做的那样待在利基里,但现在不再管用
  • 要么试图通用,提供一切任何东西,能撑一会儿,直到客户意识到不如直接用 SaaS 本身所用模型的制造者,即大科技和大 AI 公司,他们要么免费要么便宜得多

所有人被大科技和大 AI 将死了。

这将是本世纪的创造性毁灭。

但会有一些幸存者,然后大量新公司,不过它们会是什么样没人知道!


@marclou:People from 101 countries have joined!

🇮🇳 India (21%)
🇺🇸 United States (11%)
🇫🇷 France (7%)
🇬🇧 United Kingdom (4%)
🇨🇦 Canada (4%)
🇪🇸 Spain (3%)
🇩🇪 Germany (3%)
🇮🇹 Italy (3%)
🇵🇰 Pakistan (2%)
🇵🇱 Poland (2%)
🌍 Other countries (40%)

We’ve got builders joining from Madagascar, Mongolia, and even 10 from Nepal 🤗

来自 101 个国家的人加入了!

印度 21%、美国 11%、法国 7%、英国 4%、加拿大 4%、西班牙 3%、德国 3%、意大利 3%、巴基斯坦 2%、波兰 2%、其他国家 40%。

有来自马达加斯加、蒙古的建设者,甚至尼泊尔有 10 人。


@marclou:24h into the $100K Hackathon. People are building:

🎬 Video, photo & design tools (17%)
📣 Marketing & SEO tools (16%)
🛍️ Business & ecommerce tools (14%)
🤖 AI agents & dev tools (9%)
⚡ Productivity apps (9%)
🎮 Games & entertainment (8%)
🧠 Health & learning apps (6%)
🎁 Personalized gifts & Christmas products (6%)
🗂️ Directories & marketplaces (5%)
💡 Other (10%)

n = 10% of the 4,334 participants who've already shipped something.

$10 万黑客松进行 24 小时。人们在构建:

视频、照片与设计工具 17%、营销与 SEO 工具 16%、商业与电商工具 14%、AI 代理与开发工具 9%、生产力应用 9%、游戏与娱乐 8%、健康与学习应用 6%、个性化礼物与圣诞产品 6%、目录与市场 5%、其他 10%。

n = 已交付东西的 4334 名参与者中的 10%。


@arvidkahl:Whenever my agents do some intense debugging (like today, where I am shaving off 50% CPU use from my RSS parsing logic due to resource contention), I ask it to just visualize it to me.

It even cites benchmarking runs and strace logs.

AI is a great teacher if you care to ask.

每当我的代理做高强度调试时(比如今天因资源争用从 RSS 解析逻辑削掉 50% CPU 使用),我就让它可视化给我看。

它甚至引用基准测试运行和 strace 日志。

如果你在乎去问,AI 是很好的老师。

[图片:调试可视化截图]


@arvidkahl:The trick that worked for me is to have ANOTHER AI do the grading. Codex reviews Claude Code's work, and vice versa. There's something in these models that, if you tell them they're reviewing a competing models output, makes them VICIOUS and very focused in their review.

对我有效的技巧是让另一个 AI 做评分。Codex 审查 Claude Code 的工作,反之亦然。这些模型里有某种东西,如果你告诉它们正在审查竞争模型的输出,它们会变得凶狠且非常专注于审查。


@swyx:devin bots are multi-agent teams done right:

  • always on, proactive (but tuned for ROI)
  • own outcomes not tasks (but can collab w others)
  • manages teams of subagents (and reviews itself)
  • available everywhere (call text slack whatever hello @interaction)
  • consistent reporting thru interactive artifacts aka devinapps not long chats (HTML is all you need™ @trq212)

this is the preview that @walden_yan gave at AIE. i've been trialing them for a week and i've been able to hand over my AEO, QA, cross-platform app porting, personal admin automation, and video editing work (also simultaneously moving from local agent worktree hell to cloud because ofc this is all cloud based thank god)

hyped concepts this replaces:

  • /goal
  • "ultracode" / "ultra mode"
  • "loop engineering"
  • "agent to agent"
  • "agent swarms"
  • AI employees (see below)

my main aha moment is that you don't want "AI Employees", we do not have agi yet and you will just fail if you overpromise/overexpect. what you want is the "just" the next level of abstraction above single tasks, and the overwhelming consensus I've had talking to everyone at the frontier of agents (yes incl cog competitors) is agent managers that own outcomes.

incidentally, as an employer of humans it's also how i distinguish between someone who "has agency" (owns a problem area, finds issues and researches/brings solutions not problems) and who doesn't (only does precisely what i asked, not proactive, doesn't come up with suggestions i didn't think about). guess which one i value more lol.

if youre a startup, reminder that Devin for Startups offers 65k in credits. apply via clanker

Devin bots 是正确的多智能体团队:

  • 始终开启、主动(但针对 ROI 调优)
  • 拥有结果而非任务(但可与他人协作)
  • 管理子代理团队(并自我审查)
  • 随处可用(打电话、发短信、Slack 什么的,hello @interaction)
  • 通过交互式产物即 devinapps 一致报告,而非长聊天(HTML 就够了)

这是 @walden_yan 在 AIE 给的预览。我试用一周了,能把 AEO、QA、跨平台应用移植、个人行政自动化和视频编辑工作交出去(同时从本地代理工作树地狱移到云端,因为当然全是云端)。

这取代的炒作概念:

  • /goal
  • "ultracode" / "ultra mode"
  • "loop engineering"
  • "agent to agent"
  • "agent swarms"
  • AI 员工

我主要的顿悟是你不想要“AI 员工”,我们还没有 AGI,过度承诺/期望只会失败。你想要的只是单任务之上的下一层抽象,我跟所有代理前沿的人(包括竞品)聊后的压倒性共识是拥有结果的代理管理者。

顺便作为人类雇主,这也是我区分“有 agency”的人(拥有问题领域、发现问题并研究/带来解决方案而非问题)和没有的人(只精确做我问的、不主动、不提我没想到的建议)的方式。猜猜我更看重哪个。

如果你是初创,提醒 Devin for Startups 提供 65k 积分。通过 clanker 申请。

[视频:Devin Bots 演示]


@jasonfried:Thanks Neeraj. We wanted to get the MCP right, sorry it took a bit longer to launch that you'd like. But yes, we have it now, as well as a CLI. https://basecamp.com/ai

谢谢 Neeraj。我们想把 MCP 做对,抱歉比你想要的晚了点发布。但现在有了,还有 CLI。

[链接:Basecamp AI 页面]


@jasonfried:We have dozens doing exactly that. Our account has some extra stuff as we build this out for others, but you can host an agent locally, or use a web-based agent, and just invite it as a normal user. Give it an email, invite that email to Basecamp, and they're in.

我们有几十个在这么做。我们账户有一些额外东西因为我们在为别人构建这个,但你可以本地托管代理,或用网页代理,然后作为普通用户邀请它。给它一个邮箱,邀请那个邮箱到 Basecamp,它就进去了。


@dhh:Omarchy on a 5" mini machine is a great fit for tiling!

Omarchy 在 5 英寸迷你机器上非常适合平铺!


@dhh:Local models locally, local models remotely, local models everywhere!

本地模型本地跑,本地模型远程跑,本地模型到处跑!

[图片:相关截图]