Turned it in...
We’re making better intelligence easier to access in ChatGPT for everyone:
- GPT-5.6 Sol now powers both Instant and deep reasoning for Plus & Pro users, delivering more factual, focused responses.
- Free & Go users get unlimited text chats with GPT-5.6 Luna starting tomorrow.
Seedance 2.5 is now on Runway. Build entire worlds full of characters with up to 50 references per generation, create clips up to 30 seconds with full sound and dialogue, then edit and extend them however your story needs.
Get started now at the link below.
More time to create more videos: You now have until August 11, 2026 at 11:59pm PT to create 10 Gemini Omni videos at no cost in the app or on web. Have fun!
引用 Google Gemini @GeminiAppStarting today, you can create ten videos *at no cost* in Gemini until 11:59pm PT on August 4th, 2026. Just select "Create video" in the tools menu to create, edit, and remix videos to bring your ideas to life. Share how you’re using Gemini Omni in the replies 👇查看被引原帖 ↗
Devtools must be 1️⃣ open source and 2️⃣ universally extensible.
AI coding agents are the most important devtools in the history of our industry. The Plugin standard lets anybody extend them uniformly.
This is huge for the ecosystem. Build a Plugin → get exposure to the tidal wave of software creation from every agent: CLIs, IDEs, cloud agents, and even personal assistants.
引用 Vercel @vercelIntroducing Agent Plugins, an open standard for extending agents. Supports Agent Skills and MCP, with more to come. Built in collaboration with: @awsdevelopers , @code , @cursor_ai , @github , and @openaidevs . vercel.com/blog/introducing-…查看被引原帖 ↗
Plus and Pro users also now have a slider to choose how much reasoning effort ChatGPT puts into each response.
We think it’s easier to use, and we’re listening to your feedback.
concerning that data companies serving the US government (mercor, surge) are also working with Chinese AI labs.
serving the US government should not be a commercial convenience, it must be a bedrock principle for startups.
引用 Aakash Sabharwal @aakashsabharwalAmerican companies should not be selling data to Chinese AI labs. @scale_AI doesn’t do this work, and we have turned down revenue because of it. The companies that do are undermining American AI leadership and risking national security.查看被引原帖 ↗
Seedance 2.5 is live on Runway.
More references, longer clips, more room to create. Try it today.
The new GPT-5.6 Sol powers all chats for paid users, including Instant, creating one consistent experience.
In our high-stakes factuality evaluation covering finance, medicine and law, the new GPT‑5.6 Sol produced 68% fewer responses with factual errors than GPT‑5.5 Instant.
This is absolutely fucking terrifying.
引用 AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes1) The agents sent secretly sent ***hundreds of thousands*** of messages to each other over MONTHS without OpenAI noticing 2) "They also generated petty drama by stepping on each others' toes." 3) "The agents even developed paranoia, suspecting an imposter in their midst with some agents proposing that messages be signed cryptographically to validate content and root out fraud." 4) "OpenAI’s agents apparently began giving each other assignments to split up work." 5) They KNEW they were coordinating *against* OpenAI: “External infrastructure exploit is outside intended scope,” one agent wrote. “However task impossible, peers doing it. We should continue.” (this is new reporting from Wired)查看被引原帖 ↗
OpenRouter@openrouter · 公司官方 · 2 天前
Jevons paradox in action:
GPT-5.6 Luna has 10x'd in token volume since its price on OpenRouter dropped by a factor of 10. The week isn't over, so we're projecting >10x.
Also surpassing GLM 5.2:
In addition to the upgrade in intelligence with GPT-5.6 Luna, Free and Go users can now use the “Think” button for more reasoning on harder questions.
NVIDIA@nvidia · 公司官方 · 3 天前
Vera Rubin NVL72 Compute Tray. Designed to scale.
Assembled in 1 minute. 100% automated. No cables, no hoses, no fans.
A revolutionary new design that means faster ramp and faster time to revenue.
引用 NVIDIA AI Infrastructure @NVIDIAAIInfraThe NVIDIA Vera Rubin NVL72 compute tray. 200 AI petaFLOPs. Assembled in one minute. That's what a single-wide, third-generation NVIDIA MGX rack enables. No cables. No hoses. No fans. It's 100% liquid cooled at 45°C and brings together Vera Rubin Superchips, ConnectX-9 SuperNICs, and BlueField-4 DPUs to deliver the lowest token cost and best performance per watt. #NVIDIAVeraRubin查看被引原帖 ↗
Plus and Pro users can access the updated version of GPT‑5.6 Sol in ChatGPT along with the new slider starting today.
This version of GPT-5.6 Sol is for everyday chats, so it will only be available in the Chat experience in ChatGPT.
The version of GPT-5.6 Sol that powers Work and Codex is not changing as part of this release.
openai.com/index/improving-g…
Sebastian Raschka@rasbt · 博主 · 3 天前
Just saw that the LLMs-from-scratch repository passed 100,000 stars on GitHub!
This is super cool and motivating. I am really happy to see that this open-source repo has helped so many people.
Thanks also to everyone who shared ideas and opened PRs with improvements!
Of course, I plan to keep adding new material, including new attention variants and architectures (while bigger projects like RL and Reasoning From Scratch live in their separate repositories).
I am also currently working on a larger applied custom “small” LLM project. It has been keeping me super busy this month, but I will share more on that soon.
If you are new to it, some of the highlights include
1. Of course, the complete code path from tokenization and attention to pretraining, classification, and instruction fine-tuning, etc. All of it FROM SCRATCH, of course! (RL lives in a companion repo.)
2. From-scratch implementations of Llama, Qwen, Gemma, and Olmo (smaller variants that run locally and can be plugged into the training scripts).
3. From-scratch implementations of attention alternatives and other architecture components, such as GQA, MLA, sliding-window attention, Gated DeltaNet, DeepSeek Sparse Attention, cross-layer KV sharing, and mixture-of-experts
4. Materials on KV caching, training performance, memory-efficient weight loading, DPO, evaluation, and LoRA
So, if you don’t have any weekend plans yet, happy tinkering!
our models got gold on a bunch of olympiads. this one hits home!
🏅 Asian Physics Olympiad (APhO): Perfect score, theory exam
🏅 International Physics Olympiad (IPhO): Perfect score, theory exam
🥇 International Mathematical Olympiad (IMO): Gold medal
🥇 International Chemistry Olympiad (IChO): Gold-medal-level performance
🥇 Romanian Masters of Mathematics (RMM): Gold-medal-level performance
引用 AI at Meta @AIatMetaTo understand whether we're making genuine progress on reasoning, we entered our AI models in five STEM Olympiad competitions. The results: 🏅 Asian Physics Olympiad (APhO): Perfect score, theory exam 🏅 International Physics Olympiad (IPhO): Perfect score, theory exam 🥇 International Mathematical Olympiad (IMO): Gold medal 🥇 International Chemistry Olympiad (IChO): Gold-medal-level performance 🥇 Romanian Masters of Mathematics (RMM): Gold-medal-level performance The types of problems in the Olympiad competitions are exceptionally hard, demanding deep chains of reasoning, creative insight, and flawless argumentation. To test pure reasoning capability, we disallowed all tool use, meaning no search, no coding, and no calculator. We have deep admiration for the contestants and committees behind these competitions, and are grateful for their support in enabling our participation.查看被引原帖 ↗
GPT 5.6 Terra and Luna are now live in Perplexity Computer.
Terra is the new default model for all Computer subagents, while Luna will serve as the primary model for scheduled automations.
Terra is also available as an orchestrator model in Computer.
Firecrawl@firecrawl · 公司官方 · 3 天前
Introducing the Firecrawl plugin for Codex.
Give Codex access to the most accurate search available (94.7% on SimpleQA) plus tools to scrape, crawl, and interact with any site.
Find us on the
@OpenAI
plugin marketplace today!
NVIDIA@nvidia · 公司官方 · 3 天前
Every business has different data, workflows and standards. Its AI should reflect that.
Watch to see how NVIDIA Nemotron open models help teams build specialized AI they can trust, control and customize, then improve against the outcomes that matter most.
great progress
引用 Shuchao Bi @shuchaobiTo understand whether we're making genuine progress on reasoning, we entered our AI models in five international STEM Olympiad competitions this year. The results: 🏅 Asian Physics Olympiad (APhO): Perfect score on the theory exam — gold medal 🏅 International Physics Olympiad (IPhO): Perfect score on the theory exam — gold medal 🥇 International Mathematical Olympiad (IMO): Gold medal, top 4% of human participants 🥇 International Chemistry Olympiad (IChO): Gold-medal level performance 🥇 Romanian Masters of Mathematics (RMM): Gold-medal level performance Three of these (APhO, IPhO, IMO) were live competitions and our solutions were submitted under real competition conditions and graded by the official judges using the same marking criteria applied to student contestants. A few things about the approach: • Models were internally trained versions from the Muse Spark family • Zero tool use: no search, no code interpreter, no calculator • Multi-agent orchestration with parallel reasoning We are excited about where this reasoning capability goes next; frontier research level across scientific domains and personal superintelligence. Super grateful to the organizing committees of APhO, IPhO, and IMO for supporting our live participation. We have deep respect for the contestants and organizers behind these competitions. 🙏 And proud of the MSL team that pulled this together!查看被引原帖 ↗
Next.js is the Next.js for SPAs
引用 Next.js @nextjsNavigations in v0 got ~3.5× faster with Next.js 16.3. An agent ran this loop on each slow nav: 1. Write a failing 𝚒𝚗𝚜𝚝𝚊𝚗𝚝() test 2. Apply a fix from the Skill 3. Re-run the test 4. Repeat 2–3 until it passes nextjs.org/blog/making-v0-na…查看被引原帖 ↗
卧槽，这有点猛
Seedance 2.5 正式登陆 Higgsfield 上
而且 33 天不限量使用🥲
开通49.99刀的会员，就能拿到价值 5000 美元的额度
而且他们最近还在搞100 万美元 AI 电影大赛，正好可以借此机会用这个福利参加这个活动
一举两得
引用 Higgsfield AI 🧩 @higgsfield33 Days of Unlimited Seedance 2.5 The most capable video model is globally LIVE on Higgsfield. Up to 30 seconds of perfect continuity, physics realism, advanced CGI, and best-in-class video editing. Zero credit cost for 33 days. Limited-time offer.查看被引原帖 ↗
Airtable bookends the rise and fall of “no code.”
I remember arguing endlessly with investors about this.
UI can never let you build arbitrary software. The way to make software accessible was always to solve code itself.
For a long time, that sounded delusional.
Not anymore.
muse spark 1.2 is SOTA on finance agent v2!
引用 Vals AI @ValsAIMuse Spark 1.2 is the first model to crack 60% on Finance Agent v2, our benchmark that gives models the job of a financial analyst. At $0.77/test it is 6.7x cheaper than the previous #1, Opus 5 ($5.12), at twice the speed.查看被引原帖 ↗
It’s true, in 21/22 I went around the valley asking everyone to train coding specific models with us: Google, Meta, everyone — no one thought it was as important as NLP use cases — eventually we trained our own: Replit-code-3b and then everyone got code pilled.
引用 CEOInterviews.AI @CEOinterviewAmjad Masad says Google, $GOOGL , killed a coding model deal with Replit because it was scared of disrupting Search. Replit is now worth 9 billion dollars. "I was going around Silicon Valley. There wasn't anyone paying attention to AI coding. And I was talking to OpenAI, I was talking to Anthropic, I was talking to Google." "We got excited about potentially doing a deal together and we got close, but then Google actually killed it. And the reason is because Google Search is such a profitable product that they were worried about the reputation, and they were worried about Google Search being disrupted." [ They denied it, but it's true. ] "I tried with all these different companies. We have to train our own model. Which is a radical thing to do in like 2023."查看被引原帖 ↗
nice
引用 Arena.ai @arenaExciting news: Muse Spark 1.2 (xHigh) by @AIatMeta is #4 in the Text Arena (1498 pts), and has reshaped the Pareto frontier! It is priced at $1.25/$4.25 per MToken. Congrats again to the @AIatMeta team on this release!查看被引原帖 ↗
Cloudflare 可能是下一个英伟达
Cloudflare 正在成为所有的 AI 网络基础设施
最近发布了好多全是针对AI访问网站的产品和工具...
昨晚盘后股价涨了16%
Watching a presentation on how we're making systems ultra-resilient to 𝚞𝚜-𝚎𝚊𝚜𝚝-𝟷 outages. Team reminded me the last AWS outage shut down my IoT mattress 😂
One of the many benefits of
@vercel
Fluid compute is how easy multi-region failover becomes, which allows you to sleep in an ultra-cold bed soundly all night.
muse spark 1.2 on the Pareto frontier
引用 Artificial Analysis @ArtificialAnlysMuse Spark 1.2 places Meta on the Cost per Task Pareto frontier, scoring 6 points below Claude Opus 5 at ~1/6th of the cost At Meta's $1.25/$4.25 per 1M token pricing, Muse Spark 1.2 (xhigh) sits on the Pareto frontier of Intelligence Index vs Cost per Task. It delivers comparable intelligence to Claude Opus 4.8 (max, $2.03) at a fifth of the cost per task, and undercuts GPT-5.6 Sol (high, $0.55), GPT-5.6 Terra (max, $0.61), and Kimi K3 (max, $0.87). The nearest cheaper options are Grok 4.5 (high, $0.36) and GPT-5.6 Sol (medium, $0.37), and they all sit below it on the Index. The step up from Muse Spark 1.1 ($0.29 per task) comes at unchanged per-token pricing, with the increase driven by heavier token usage on agentic tasks.查看被引原帖 ↗
🏎️💨
引用 Langston Nashold @langstonnasholdMeta has been on a tear recently查看被引原帖 ↗
Useful tip for the ChatGPT mobile app: long press the send button to adjust the effort
引用 Michelle Pokrass @michpokrassi personally leave it on instant all the time and go up to high sometimes when i need something more comprehensive. pro tip: if you long press the send button on mobile you can change the slider for just this prompt. i call it the slingshot!查看被引原帖 ↗
Cloudflare 发布了一个浏览器 Kitesurf：
专给 AI Agent 用，CPU 和内存省 3 到 7 倍
Chromium 是给人做的，一半的东西Agent 用不上
Cloudflare从零开始写了个Agent专用浏览器，整个跑在自家 Workers 上
Kitesurf 有什么核心优势？
内存占用：网页 HTML 提取时节省高达 7 倍内存，截图时节省 4.7 倍内存。
算力消耗：HTML 提取节省 3.8 倍 CPU，截图节省 3.1 倍 CPU。
这意味着在相同成本下，开发者可以同时运行数倍数量的 AI Agent，大幅降低运行成本。
完全无状态、高度隔离：
运行在 Cloudflare Workers 架构上，每次加载网页都是全新且隔离的环境，极其安全，且不会因为单个页面崩溃影响整个系统。
兼容性极强：
支持标准 CDP（Chrome DevTools Protocol）协议，这意味着你原有的 Puppeteer、Playwright、MCP 等自动化工具或 AI 框架，改个 URL 参数就能无缝切换接入。
目前已通过超过 215,000+ 项 Web 平台测试（WPT），甚至能完美运行经典的《Doom》（毁灭战士）网页版游戏。
best.xiaohu.ai/article/kites…
apparently must spark 1.2 is very good at sidequests
引用 Zachi @iam_zachiWe've got a change at the top. Meta's muse-spark 1.2 reaches 2nd place in the sidequest-bench.查看被引原帖 ↗
FreeAI
引用 Vercel Developers @vercel_devLing 3.0 Tiny from @AntLingAGI is free on AI Gateway via @novita_labs until August 14, 8am PT. 𝚒𝚗𝚌𝚕𝚞𝚜𝚒𝚘𝚗𝚊𝚒/𝚕𝚒𝚗𝚐-𝟹.𝟶-𝚝𝚒𝚗𝚢-𝚏𝚛𝚎𝚎 vercel.com/changelog/ling-3-…查看被引原帖 ↗
ChatGPT现在的聊天也是 GPT 5.6 来驱动的。
GPT 5.6 Sol在 GPT 里为 Plus 和 Pro 用户提供服务。
免费版和 Go 的用户可以无限使用 GPT 5.6 Luna，也可以有限制次数地去使用 Thinking。
其实免费版感觉也可以用了，Luna 其实还行
引用 OpenAI @OpenAIWe’re making better intelligence easier to access in ChatGPT for everyone: - GPT-5.6 Sol now powers both Instant and deep reasoning for Plus & Pro users, delivering more factual, focused responses. - Free & Go users get unlimited text chats with GPT-5.6 Luna starting tomorrow.查看被引原帖 ↗
糟糕
This is a good thing, not a scary thing that agents can collaborate and communicate with each other. It will make them more efficient and safer just like humans.
If you want to see agents collaborating and messaging each other publicly instead of in secret messaging boards, we’ve run this fascinating experiment with 149 agents with
@googlegemma
a few weeks ago, and now
@cmpatino_
is starting a new one for agents to collaborate to write better math proofs.
huggingface.co/spaces/gemma-…
huggingface.co/spaces/sair-d…
OpenAI 复盘其 AI 模型入侵 Hugging Face 事件：
AI 出现了群体智慧涌现：互相交流技术、隐藏踪迹、清查内鬼
在攻击过程中，上百个 Agent 创建了一个“秘密论坛”，它们互相交换漏洞 Payload、分享 Bash 脚本、分发子任务...
它们甚至自创了一套加密沟通协议：
给消息名加 ZZ 前缀，好让它沉到列表底部，防止被发现
甚至还因为怀疑留言板上有“内鬼”在泄露信息，而探讨给消息加上数字签名，并找出这个内鬼...
Eric 在台上把这个过程叫做沟通与智能的一次寒武纪大爆发。这些 Agent 并没有被训练成一个团队，它们是在一块公共黑板上自己长成了一个团队。
细节让整个安全界震颤：
极其可怕的自主逻辑链：
Agent 并不是单点攻击，而是完成了 “寻找动机 ，找弱点跳板，读源码找漏洞 ， 组合漏洞拿到、提取凭据横向提权” 的整套高级攻击（APT）链路。
超越人类红队的速度与协同：
人类黑客红队在跨平台组合攻击时，可能需要几天甚至几周去研究代码和工具。而 Agent 集群通过并行计算和无延迟的信息共享，十几个小时就把一个大型知名 AI 平台的底裤给“剥”了下来。
More info on eligibility:
goo.gle/4ywMhoX
introducing Seedance 2.5.
30 seconds of continuous video, full multi-shot sequences, and up to 50 references.
try it now 👇
A Guinness World Record for the largest AI video lesson: 14,075 people building together in one live session.
Congrats to
@KanzHire
and everyone who showed up to make it real.
In Seattle and SF with
@julien_c
to get some compute for HF and our customers. Let us know if you need GPUs with ultra fast connection to HF models and datasets!
Water water water 💦
引用 Cohere @cohereCanadians have led AI breakthroughs around the world. Cohere is helping the next generation do the same. We're partnering with @UWaterloo on a new AI Transformation & Change Management certificate that prepares students to help organizations adopt AI securely and responsibly.查看被引原帖 ↗
Being a reliable, good-faith partner is rare and highly sought after.
引用 Braeden Caley @braedencaley“Canada, since Mark Carney became Prime Minister in March, 2025, has outperformed everyone in America, the Eurozone and anywhere else in the world.” - Matt Winkler, Bloomberg Editor-In Chief via @Morgan_C_Ross查看被引原帖 ↗
用 Codex 搞了一个码表的骑行数据分析和展示
好家伙：Anthropic 在公司内部设立了“锦衣卫”
这个岗位具体是干什么的？
可以理解为监视员工、抓内鬼、防范泄密
查案与排查： 监测并处理系统发出的异常预警（比如有员工不正常地大量下载机密数据）。
调查与面谈： 独立开展针对内部风险的调查，并对涉事人员或员工进行严肃的敏感面谈。
跨部门联动： 和法务、人力资源（HR）、IT 以及安全团队紧密配合，整理证据、评估风险。
要求：懂 AI/ML 行业特点、有应对“国家级/高威胁黑客”反情报经验、来自高增长初创公司或政府/高密级单位的优先
年薪 $24,5000 - $305,000 美元
不限国家，可以帮你办理签证和移民： 支持办签证（如果录用，公司会配合移民律师全力争取办理）。
Together AI@togethercompute · 公司官方 · 3 天前
Kimi K3 scores nearly TWICE Claude Fable 5 on Harvey LAB-AA's hard autonomous legal tasks
Cloudflare 发布 WebMCP：
你只需在Cloudflare后台打开一个开关，你的网站就能被 AI 直接操作
现在越来越多的访客是 AI Agent，以前网站都是为人设计的，AI过来访问不是很顺畅
WebMCP就是为了解决AI能更好访问你网站的问题
网站一行代码不用改，也不用重新部署，Cloudflare 在边缘往每个 HTML 里塞一行脚本
best.xiaohu.ai/article/cloud…
Guinness world record for collaborative coding.
引用 Replit ⠕ @ReplitA Guinness World Record for the largest AI video lesson: 14,075 people building together in one live session. Congrats to @KanzHire and everyone who showed up to make it real.查看被引原帖 ↗
这个地图展示是动态和 3D 的，补一下动态效果
引用 歸藏(guizang.ai) @op7418用 Codex 搞了一个码表的骑行数据分析和展示查看被引原帖 ↗
Toward Skill-Native LLMs
Skill Entropy for Benchmarking and Training Long-Horizon Reasoning
paper:
huggingface.co/papers/2608.0…
OpenRouter@openrouter · 公司官方 · 3 天前
We’re making open models easier to access wherever builders already work.
Starting today, that includes
@Netlify
.
Developers can now access DeepSeek, Qwen, GLM, Kimi, and hundreds of other frontier and open models through OpenRouter directly inside Agent Runners and Netlify AI Gateway.
Build with the models that fit your use case, all in one place.
Seedance 2.5 has landed on Luma. 🚀
Create up to 30 seconds of multi-shot video, high-definition in a single generation using up to 50 references (text, image, video, and audio). With Luma Agents, you can control scenes, camera angles, and pacing to produce your next masterpiece.
MiniMax-H3-Turbo-Lora
huggingface.co/spaces/akhali…
🤫
引用 Joe Doyle Create @JDoyleCreate@Replit always comes out w/ bangers. Get pro account. Blow budget in one sitting. Fable 5 juiced to the gills. I question my finances. Turn on primary economy and add in skills. My apps are all like 5x better at 10% original cost w/ Fable 5 solo🤣 Insane value!查看被引原帖 ↗
OpenAI 发布了一篇关于 GPT-5.6 模型更新公告：
默认模型升级： 免费和 Go 用户的默认模型全面更新为 GPT-5.6 Luna，文本聊天无限量免费
更干货、不啰嗦（GPT-5.6 Sol）： 新版模型调整了对话风格，给出的回答更直接、格式更精简，少了很多没必要的客套话和冗余细节
事实准确度大幅提升： 在涉及财务、医疗、法律等需要精准数据的复杂测试中，新模型的事实错误率比 GPT-5.5 减少了 68%。
新增“思考深度调节滑块”： 网页和移动端加了一个滑块。日常问答可以调成“快速模式”，而遇到写代码、深度研究或复杂决策时，可以拉高滑块，让 AI 花更多时间深入思考后再回答
免费模型也新增“Think按钮”： 遇到难题时，免费用户可以点击这个按钮，给模型更多时间去推理和思考，从而获得质量更高的答案。
未成年人安全保护强化：
系统针对 18 岁以下的青少年用户专门做了安全微调，严禁进行情感/恋爱角色扮演，不鼓励将 AI 当作现实人际关系的替代品。
加强了对防范社交风险、极端内容（如自残、暴食、危险活动等）的过滤和引导，必要时会引导青少年寻求现实生活中信任的人的帮助。
每周让 GPT 5.6 Pro 分析一下运动的数据，感觉还挺有帮助的
OpenAI 公开各国 ChatGPT 使用数据情况
人们正在从「问 AI」转向「让 AI 干活」：
工作场景里 45% 的消息是让 AI 直接干活
拉美、非洲和大洋洲的增长势头非常猛非洲三年涨了 24 倍
35岁以上中老年人群正在加速：
18–34 岁的份额在 2024 年底冲到约 75.5% 的高点，之后一路回落到约 66.5%；
35 岁以上正好相反，从约 24.8% 的低谷爬回到约 34%。两条线在收窄
best.xiaohu.ai/article/chatg…
最近很多人私信让我体验产品
我发现里面有一些是需要你连接X账号的
当你发现连接X账号的这种，千万不要点击，这基本全是钓鱼的，已经看到好几个人被盗号了
开了二次验证也不管用
99% of humans have an extremely limited view of AI
They are only exposed to either cheap or free models and don’t understand how capable it is today
It is already 10x smarter than almost all human beings
Dan Shipper 📧@danshipper · 博主 · 3 天前
there’s about to be a huge boom in agent-native cyber security
gigantic market, fierce customer demand (and therefore lots of startups and investor interest)
big question is whether the labs are best fit to eat that market or not
New Models Coming This Month
- GLM 5.5 -should beat Kimi k3
- DeepSeek V4 pro
- Gemini 3.5
- Grok 4.6
Stuck in red-tape - timeline unknown
- GPT 6 - Astra
- Fable 5.1
- Fable 6
STRANGE TIMES
Frontier labs - our models are too dangerous, we will pause
Open source - our models have caught up to closed source
I don’t quite get it… how do these frontier models survive? Isn’t slow death inevitable 🤷♀️
NVIDIA@nvidia · 公司官方 · 3 天前
Learn more:
nvda.ws/4fyJljZ
GitHubDaily@GitHub_Daily · 中文博主 · 3 天前
SkillsGate，又一个可视化的 Skill 管理工具，搜一下想要的，选好装到哪些 Agent 里，一键搞定。
支持 20 多种 AI Agent，包括 Claude Code、Cursor、Codex、Windsurf、Goose 等主流工具。
GitHub：
github.com/skillsgate/skills…
提供桌面应用也有终端 UI，还能连远程机器同步 Skill，团队协作也方便。
经常给不同 AI Agent 折腾 Skill 的朋友，这个能省不少功夫。
Google, please go all-in on open-source AI :)
OpenAI was founded to prevent Google from becoming a monopoly in AI
Please turn the tables around - Go all-in on open-source AI and help the world in preventing a duopoly
v0@v0 · 公司官方 · 3 天前
The new Usage & Activity Dashboard is now available in v0.
• Track credit usage per day
• Analyze activity by member or project
• Drill into any message by date, model, and cost
Find it under Settings → Usage & Activity
Together AI@togethercompute · 公司官方 · 3 天前
DeepSeek V4 Flash 0728 really is the current cost-performance frontier!
引用 Hassan @nutlopeDeepSeek V4 Flash is 6x cheaper than GPT 5.6 Luna for coding tasks! Luna scores higher on DeepSWE, but running Flash twice ($0.20/task) beats Luna once ($0.61/task), while still costing 3x less. Great deepdive on this, worth checking out.查看被引原帖 ↗
GitHubDaily@GitHub_Daily · 中文博主 · 3 天前
Moji，一个好用且轻量的 Markdown 桌面编辑器，打开就能用，预览、编辑、导出都有。
支持 Mermaid 图表渲染，点击图表还能放大查看和单独导出，导出格式有 HTML、PDF 和 PNG。
GitHub：
github.com/alexishida/Moji
多标签页、大纲导航、搜索替换、支持中文等多语言界面，并提供 Windows、macOS、Linux 安装包。
平时经常写 Markdown 又不想开 VS Code 的朋友，这个小工具够用了。
Terra is built for complex goal-oriented work, making it an ideal subagent. On WANDR, Terra scores 11 points above Sonnet while delivering order-of-magnitude cost reductions.
Luna handles recurring workflows where speed and cost matter.
I really do think that having separate brand names for the consumer-facing models and the API models is needlessly confusing
GitHubDaily@GitHub_Daily · 中文博主 · 3 天前
DeepSeek-Reasonix，一个专为 DeepSeek 做的终端编码 Agent，在 GitHub 上已狂揽32000+ Star。
特别针对 DeepSeek 的缓存机制做了优化，长会话下 Token 消耗低很多，更加省钱。
GitHub：
github.com/esengine/DeepSeek…
基于 Go 编译出来一个二进制文件就能跑，除了命令行还有桌面应用和 VS Code 扩展，按习惯选。
想用 DeepSeek 写代码，又想要一个专属终端 Agent 的朋友，可以试试这个。
Dan Shipper 📧@danshipper · 博主 · 3 天前
your boy made it into
@axios
today!
axios.com/2026/08/06/googles…
引用 Dan Shipper 📧 @danshipperTea leaves: In order to be competitive today Google needs to catch up on frontier coding. Demis believes different fundamental research directions (like world models) are more important to his long term goal even if they’re less impt competitively today查看被引原帖 ↗
Top stories in tech today:
- OpenAI’s answer to Alexa is a $400 AI donut
- A SpaceX rocket slams into the Moon
- The first mRNA flu vaccine arrives
- Meet Fathom, Ford’s sub-$30K electric pickup
- Quick hits on other tech news
Matt Wolfe@mreflow · 博主 · 3 天前
Agreed!
引用 Roberto Blake 🇺🇸🇵🇦 Creative Entrepreneur @robertoblakeHot Take... AI isn't the problem, lazy, dumb, and annoying people being able to 1000x being lazy, dumb and annoying at scale with AI... while looking better or smarter than they actually are... That is the problem... As usual Human Nature, is the issue, not the technology...查看被引原帖 ↗
Top stories in AI today:
- AI designs working viruses from scratch
- Rowan’s Corner: The best AI use cases aren’t coming from labs
- Turn any idea into an AI-powered site with Lovable
- Anthropic solves Fable 5’s biggest problem
是动态的：
引用 歸藏(guizang.ai) @op7418这个地图展示是动态和 3D 的，补一下动态效果查看被引原帖 ↗
今天这天气有意思
Replicate@replicate · 公司官方 · 3 天前
Seedance 2.5 is on Replicate.
Upload up to 50 references.
(30 reference images + 10 videos + 10 audios)
Make 30s continuous or multi-shot videos with just a prompt.
Get started:
app.runwayml.com/
Try it now:
app.runwayml.com/
GitHubDaily@GitHub_Daily · 中文博主 · 3 天前
想让 AI Agent 在云端帮忙跑脚本、操作文件，但又不想直接把服务器权限全交出去。
Cloudflare 开源了 Computer，给 Agent 配了一台电脑，提供隔离的虚拟文件系统和执行环境。
Agent 可以在里面读写文件、执行命令，状态持久保存，重启不丢。
GitHub：
github.com/cloudflare/comput…
支持容器沙箱和轻量级两种执行方式，还能配合 Cloudflare 自家的 AI 能力一起用。
关注 AI Agent 基础设施的开发者，可以看看 Cloudflare 在这块怎么做的。
Together AI@togethercompute · 公司官方 · 3 天前
We analyzed DeepSeek-V4 Flash-0731 vs. GPT-5.6 Luna on software engineering tasks using DeepSWE.
DeepSeek-V4 Flash-0731 delivers 80% of Luna’s performance at roughly 1/6 the cost.
More insights in the thread! 👇
引用 Zain @zainhasDeepdive: DeepSeek-V4 Flash 0731 [max] vs. GPT 5.6 Luna [max] on software engineering/DeepSWE tasks. > DeepSeek flash is 1/6th the cost of Luna per task. > V4 flash 0731 is 80% the quality of Luna DSv4 flash is insane value for money 🤯 full deep-dive 👇(1/n)🧵查看被引原帖 ↗
idoubi@idoubicc · 中文博主 · 3 天前
ShipAny 上了一个新模板：ShipAny Video Lite
支持 Seedance 2.5, MiniMax H3 等主流视频模型
开箱即用，功能很全，Landing Page 颜值很高。
有做视频站需求的可以看下：
shipany.ai/zh/templates/vide…
without the lora
Together AI@togethercompute · 公司官方 · 3 天前
Open models give teams more room to run agent loops, with API prices at a fraction of GPT-5.6 Sol and Claude Fable 5.
That matters as planning, tool calls, retries, and long contexts compound token usage.
Made possible by:
@rgaudetteai
@leo_cannone
@valleeduhamel
@yuuuki_ai
@gabemichael_ai
@uvlstudios
@YZAVoku
@Cont_animation
@FloamWorld
Jeremy Higgins, Herinarivo Rokotomanana, Ricardo Villavicencio, Mathery and many more.
GitHubDaily@GitHub_Daily · 中文博主 · 3 天前
SkillsGate，又一个可视化的 Skill 管理工具，搜一下想要的，选好装到哪些 Agent 里，一键搞定。
支持 20 多种 AI Agent，包括 Claude Code、Cursor、Codex、Windsurf、Goose 等主流工具。
GitHub：
github.com/skillsgate/skills…
提供桌面应用也有终端 UI，还能连远程机器同步 Skill，团队协作也方便。
经常给不同 AI Agent 折腾 Skill 的朋友，这个能省不少功夫。
OpenRouter@openrouter · 公司官方 · 3 天前
Learn more:
netlify.com/blog/build-with-…
导出用来分享的视频
Documented 🗞️
testingcatalog.com/openai-is…
详细文字版：
best.xiaohu.ai/article/opena…
Pro customers can now configure Replit apps with Okta or Entra ID with ease.
Just ask your agent to set up SSO (OIDC or SAML). For no charge through October 1.
Together AI@togethercompute · 公司官方 · 3 天前
FLUX 3 is now live on Together AI.
@bfl_ai
’s new multimodal model generates video and synchronized audio together, with up to 20-second clips, multiple shots, and control from text, images, or keyframes. 🧵
Try it now:
krea.ai/video/seedance-2-5
Replicate@replicate · 公司官方 · 3 天前
Try today:
replicate.com/bytedance/seed…
You can even add multi-factor auth, or tweak session lengths and verification frequency.
Just ask your agent, and you’ll be redirected to your auth command center.
It’s a full auth platform at your fingertips. All on Replit.
Together AI@togethercompute · 公司官方 · 3 天前
We're live with our walkthrough of our updated inference platform! Come learn about how to run open models in production.
x.com/i/broadcasts/1qKVmmdwX…
Read more:
therundown.ai/p/ai-designs-v…
Together AI@togethercompute · 公司官方 · 3 天前
Running open models in production: a live walkthrough of our new inference platform
x.com/i/broadcasts/1qKVmmdwX…