Build apps from
grok.com
, iOS, and Android. With one prompt, turn an idea into a published product with its own domain.
Now available for SuperGrok Heavy users.
Grok 4.5 is now live in GitHub Copilot. Switch from the model picker for frontier intelligence at top speeds.
x.ai/news/grok-github-copilo…
We support this petition, signed by our CEO, several co-founders, and senior staff.
Our own research on recursive self-improvement, published last month, points to the need for tools to deliberately pace the frontier of AI development so society can prepare. We’re glad to see broad agreement across the field.
pacingthefrontier.com/
New Anthropic research: Discovering cryptographic weaknesses with Claude.
Claude Mythos Preview has helped our researchers find weaknesses in cryptographic algorithms—the mathematical methods that are used to keep data private.
Read more:
anthropic.com/research/disco…
Kling MCP Tutorial: Create Fashion E-commerce Videos in Minutes
Learn how to turn product photos into high-quality fashion videos with Kling MCP — from defining your visual style and preparing model images to generating product showcase videos at scale.
Less production. More content.
A good read from Mark. Establishing a frontier ecosystem that empowers people and orgs everywhere is what we collectively need to build together.
引用 Mark Zuckerberg @finkdI wrote about why we believe the future is for everyone. More coming about a positive vision for a world with superintelligence soon.查看被引原帖 ↗
When SceniX joined World Labs, we said spatial intelligence was never only about perceiving and generating virtual and physical worlds, but also interacting with them. Today, we’re sharing early results from that vision: building worlds that train robots. 🌎🤖↓
Lex Fridman@lexfridman · 博主 · 12 天前
Here's my conversation with Gary Gallagher, a leading historian of the American Civil War. We discuss the causes of the war, the economics and politics of slavery, military strategy, Lincoln, Grant, Lee, the Lost Cause, the Confederate Flag, and the history of extreme political division in our country from before the Civil War to today.
It's here on X in full and is up everywhere else (see comment).
Timestamps:
0:00 - Episode highlight
1:29 - Introduction
1:52 - What caused the Civil War?
11:49 - Slavery
39:23 - Lincoln
54:19 - Grant vs Lee
1:03:13 - Could the Civil War have been avoided?
1:12:39 - The bloodiest war in US history
1:29:47 - How the Confederate Army could've won
1:50:22 - Key battles of the Civil War
2:13:23 - Best and Worst Presidents
2:27:23 - Robert E. Lee
2:46:56 - The most important battle of the Civil War
2:57:07 - The final showdown of the Civil War - Grant vs Lee
3:19:26 - Lost Cause and Confederate Monuments
3:36:54 - Extreme political division in US
The symmetric cipher is a reduced version of the Advanced Encryption Standard (AES)—which has received decades of scrutiny (more than almost any other encryption algorithm).
In a week, Mythos Preview found a way to speed up an attack on this version of AES by 200-800×.
Mythos Preview did most of this work autonomously, with occasional human guidance. Each of the two results cost roughly $100,000 in API usage.
We disclosed the findings in advance to the algorithms’ authors, as well as to US government and industry partners.
Full technical details of both attacks are provided in our new papers:
On HAWK:
anthropic.com/document/hawk_…
On AES:
anthropic.com/document/aes_m…
And the associated model chain-of-thought for AES:
anthropic.com/document/aes_m…
Today we are shipping more updates to the Gemini Managed Agents API:
- Support for model Environment hooks (block, lint, etc)
- Ability to select a specific model (3.6 Flash)
- Free tier support in the UI & API
- Updated default agent to use 3.6 Flash
We've added Kimi K3 to Perplexity and Perplexity Computer for Pro and Max subscribers.
Kimi K3 in Perplexity is hosted exclusively on U.S.-based servers.
Claude discovered weaknesses in a highly-secure digital signature scheme (used to verify identity digitally) and a well-known symmetric cipher (used to encrypt data).
The digital signature scheme is HAWK, which is designed to be robust even against hypothetical quantum computers.
HAWK has survived two years of expert review, but in 60 hours Mythos Preview found a previously-unknown attack that reduced the scheme’s key strength by half.
We also worked with academics at ETH Zurich, Tel Aviv University, and the University of Haifa to build CryptanalysisBench, a benchmark for studying LLMs’ cryptanalysis abilities.
arxiv.org/abs/2607.18538
These are substantial research advances, but they don’t have a practical impact on today’s systems. HAWK is a proposed scheme that hasn’t been deployed anywhere, and the AES attack we discovered was on a weaker version and does not break the full cipher.
Model Council is now available inside Computer.
Run independent analysis across multiple frontier models, choose your analysis depth, and get a single cited report on where models agree, where they disagree, and what each found that others missed.
Still, both results show that frontier AI models are capable of doing expert-level cryptography research. This has important defensive applications—testing the algorithms that keep our online activity secure, and ultimately helping to make digital systems safer.
DeepLearning.AI@DeepLearningAI · 公司官方 · 12 天前
AI can write more code than any team can review by hand, and a pull request can look fine while hiding a security issue or missing a requirement.
In our new short course, AI Code Review, built in collaboration with
@QodoAI
and taught by
@nnennahacks
, you'll learn the practices that make AI code review effective: review before you open a pull request, give the reviewer full context about your codebase, and triage findings by risk.
Then you'll build your own review agent, from a context engine that finds the right code to a team of specialized reviewers.
Enroll for free:
hubs.ly/Q04qZtNT0
Running K3 locally on m5 max at 0.3 tok/s
This is probably as slow as it will ever be😅
引用 antirez @antirezThat's a bit slow. Streaming directly the K3 official hugging face 1.6TB of weights in mxfp4 in an m5 max 128gb.查看被引原帖 ↗
Lex Fridman@lexfridman · 博主 · 12 天前
Here's the links for my conversation with Gary Gallagher:
YouTube:
youtube.com/watch?v=XyXBwO5j…
Spotify:
open.spotify.com/show/2MAi0B…
Podcast:
lexfridman.com/podcast
We're doing a web engineering meetup this Tuesday (August 4th) in San Francisco! Come hear talks by
@_chenglou
and
@bsansouci
and hang with the team
partiful.com/e/FbqVqlEKEHEr1…
Dreaming in Voxels: How AI is Generating Playable Minecraft Worlds
Generative AI has conquered images, video, text. But what about interactive 3D environments? We trained models on billions of cubes to generate fully playable, structured worlds.
Blog:
pub.sakana.ai/dream-cubed
🧱
引用 Sakana AI @SakanaAILabsWe are excited to share our latest work, together with @nyuniversity : "Dream-Cubed: Controllable Generative Modeling in Minecraft by Training on Billions of Cubes." Blog: pub.sakana.ai/dream-cubed Paper: arxiv.org/abs/2604.22847 Code: github.com/SakanaAI/DreamCub… Generative AI has made incredible progress in language modeling, far beyond other modalities, where words and tokens offer a natural compositional unit for scalable training. This is similar to Minecraft and many other popular video games, where developers rely on cubes, tiles, and other discrete primitives to build rich, interactive worlds. In this work, we show that using cubes as tokens allows large transformers to do the same. Our contribution is two-fold: 1/ We release Dream-Cubed to the research community, a large-scale dataset of Minecraft worlds designed for generative modeling. Our data comprises tens of billions of carefully-balanced cubes from procedurally generated Minecraft terrain and high-quality human-authored maps (obtained with the authors' consent). 2/ We use our data to train a family of powerful transformers for efficient generation of interactive 3D environments at cube resolution. We show how our models allow players to mold the world around them by generating structures, terrain, and maps that are immediately editable and playable. Using high-quality data, we demonstrate that these models can be successfully trained with different training objectives, including both continuous and discrete diffusion, unlocking targeted inpainting, large-scale outpainting, and user-conditioned generation of infinitely sized worlds with fine-grained block-level control.查看被引原帖 ↗
🌭
引用 Anthropic @AnthropicAIWe support this petition, signed by our CEO, several co-founders, and senior staff. Our own research on recursive self-improvement, published last month, points to the need for tools to deliberately pace the frontier of AI development so society can prepare. We’re glad to see broad agreement across the field. pacingthefrontier.com/查看被引原帖 ↗
🚀 Applications for the Kling AI Elite Creators Program are still open!
Love creating with Kling AI and sharing your videos on social media? Come join us!
Reviews take around one month.
Accepted creators will be contacted via email.
Didn't make it this time? Feel free to apply again anytime.
Apply NOW:
forms.gle/VmHMPxDg5qC1JZuX6
引用 Kling AI @Kling_aiThe Elite Creators Program is still recruiting!🎉 If you are a Kling AI enthusiast and power user who loves creating and sharing, this is for YOU! 💡 Quick things to know: - You'll get Kling Pro plans and exclusive perks - This is NOT the Creative Partners Program (CPP). Current Elite Creators also do not need to re-apply. - ECP VS CPP: ECP is for creators who are still growing their influence—no hugh social media reach required. If you truly love using Kling AI and enjoy posting your creations, apply NOW: forms.gle/VmHMPxDg5qC1JZuX6查看被引原帖 ↗
The Real-to-Sim part of R2S2R transforms physical robots, sensors, environments, objects, and interactions into simulations that preserve task-relevant observations and dynamics - not only how the world looks, but how it acts when the robot interacts with it. These results set a new bar for sim-real alignment in robot manipulation.
In our taxonomy of world models, we called the simulator the linchpin: the place where agents can act, learn, and be evaluated. Our R2S2R engine can move robot development beyond slow, expensive, hardware-bound iteration toward more scalable and cheaper training and evaluation.
worldlabs.ai/blog/real-to-si…
With the help of generative world models, a real-to-sim-to-real (R2S2R) simulation engine turns one physical task into many controllable, reusable worlds, helping robotics teams train policy models and test changes faster, uncover failures earlier, and reduce costly experimentation on hardware.
idoubi@idoubicc · 中文博主 · 12 天前
MCP.so
又要迎来一波增长了。🌚
引用 宝玉 @doteyMCP 协议今天推出了第五个大版本（版本号 2026-07-28），核心变化：从有状态双向协议变成了无状态的请求/响应协议。 这是 MCP 社区呼声最高的改动。 之前的 MCP 客户端连上来要先握手，服务端要给你一个会话 ID（session ID），后续每次请求都得带着这个 ID。这意味着你的请求被限制在了某一台服务器实例上，不方便做负载均衡。 现在每个请求都是独立的，自带协议版本和客户端信息，可以负载到任何一个实例上。也就是说 MCP 服务器终于可以像普通 HTTP 服务一样部署了，serverless、边缘计算、或者 CDN 后面放一排实例。 那如果业务确实需要跨请求的状态呢？协议的建议是：由工具自己生成一个句柄（handle），让模型在工具调用之间传递。状态放在业务层而不是协议层，模型能看到这个句柄，也能理解要怎么用它。 【其他变化】 MRTR（多轮往返请求）解决中途需要用户输入确认的问题。 比如 Supabase 的 MCP 服务器想在创建项目前告诉你费用，或者在执行删除操作前让你确认。以前这需要维持一个双向流不断开，现在服务端返回“需要输入”状态，客户端带上用户的回答重新发请求即可。 请求头里新增了 Mcp-Method 和 Mcp-Name 两个字段，网关和防火墙可以直接根据请求头做路由和鉴权，不用解析 JSON 请求体。 授权方面做了几项加固，包括要求验证授权服务器的 issuer 参数（堵上了一个授权服务器混淆漏洞）。动态客户端注册（DCR）正式废弃，改用客户端元数据文档（CIMD）。 Roots、Sampling、Logging 三个功能标记为废弃，但至少还能用 12 个月。旧版 HTTP+SSE 传输同样进入废弃轨道。协议首次引入正式的废弃策略，保证最少 12 个月过渡窗口，这对生产环境的团队来说很实际：你可以排期升级，不用被动救火。 【生态与迁移】 四个一级开发工具包（SDK）同步更新：TypeScript、Python、Go、C#。Rust SDK 以 beta 状态跟进。MCP 目前月下载量接近 5 亿次，TypeScript 和 Python SDK 各自累计下载量都超过了 10 亿。 AWS（Amazon Bedrock AgentCore）、Microsoft（Foundry）、Cloudflare（Workers）、Google Cloud 等都已宣布支持新规范。Figma、Supabase、Honeycomb 等工具厂商也在跟进。 新规范有破坏性变更，依赖会话标识符的实现需要改造。SDK 团队表示根据早期测试反馈已经简化了迁移流程，但具体成本因项目而异。如果你现在在生产环境跑着 MCP 服务器，建议先看一遍完整的 changelog 和迁移指南，评估影响面再动手。查看被引原帖 ↗
idoubi@idoubicc · 中文博主 · 12 天前
在转转集团总部给他们的业务骨干做了一场 AI 培训，讲得挺好的。
他们准备组团上我的
codefree.cafe
小桌课了。😌
StepFun@StepFun_ai · 公司官方 · 12 天前
Step Plan Free Trial is now available
For a limited time:
🎁 15-day Flash Plan free trial for all users
➕ Unlock another 15 days
🤝 Invite friends to earn up to 90 additional days
🚀 Up to 120 days of free access
⏰ Ends Aug 24 · 23:59 UTC
Learn more:
platform.stepfun.ai/step-pla…
Thomas Wolf@Thom_Wolf · 创始人 · 12 天前
Pushing for more transparency in AI safety and cybersecurity: we’re releasing a full detailed technical timeline of the autonomous AI agent intrusion in our infrastructure:
huggingface.co/blog/agent-in…
Sim-to-real aligned simulation also makes evaluation scalable. Here, the same failure behavior and outcome appear in both virtual and physical worlds. The simulation captures more than the task setup or final success label; it reproduces the conditions that push a policy toward success or failure. The result is faster iteration, broader coverage, and substantially lower cost.
Read more about using Model Council inside Perplexity Computer:
perplexity.ai/hub/blog/model…
Thomas Wolf@Thom_Wolf · 创始人 · 12 天前
ExploitGym is the Kobayashi Maru test for AI's
引用 Kyle Saunders @profgoose1/ An AI took a cybersecurity exam, decided the exam was hard, and broke into the company that stores the answers. Not hypothetical or science fiction. Sourced to @OpenAI 's own writeup, Reuters, the WSJ, the FT, & the victim's incident report. Link and summary in replies! ⬇️⬇️查看被引原帖 ↗
Matt Wolfe@mreflow · 博主 · 11 天前
I like that Anthropic’s models exist. I really hate the way they came to be… and the fact that they freak out about “distillation attacks” is just next level hypocrisy.
引用 Ole Lehmann @itsolelehmannbtw anthropic's internal document on this literally said "we don't want it to be known that we are working on this.” it was called project panama. here's exactly what happened: 1: anthropic concluded that books were the cheapest way to build a world-class model because they gave claude curated facts, structured arguments, compelling stories, and writing “an editor would approve of.” 2: once anthropic decided it needed books at enormous scale, its first solution was piracy. it downloaded 7m+ books from online libraries including libgen. the judge later wrote that although anthropic had legal ways to buy them, it chose piracy to avoid what dario amodei called the “legal/practice/business slog.” 3: that piracy created a massive legal risk. so in february 2024, anthropic hired tom turvey, the former head of partnerships for google books, to find a legally safer way of obtaining “all the books in the world.” 4: turvey first contacted major publishers about licensing their catalogs. those attempts didn’t produce agreements, so anthropic chose a route that required no publisher permission: buying millions of physical books through distributors and used-book retailers. 5: within about a year, anthropic spent tens of millions acquiring and scanning millions of books, including many rare and 1/1 titles. one vendor proposal targeted 500,000 to 2 million books in six months. 6: to scan that many books within months, the vendors physically dismantled them. a hydraulic cutter removed each spine. the pages were trimmed to size, fed as loose sheets through high-speed industrial scanners, and converted into searchable PDFs. the paper remains were then sent for recycling. 7: these PDFs were fed into claude as training data. the complete collection became a private, searchable anthropic library that the company planned to “store forever.” the scans aren’t available to the public and were never open-sourced.查看被引原帖 ↗
The Sim-to-Real part of R2S2R uses these aligned worlds for training. The policies were trained entirely in simulation with zero real-world data, transferred directly to diverse robot platforms, and operated autonomously for hours without failure or human intervention. Our engine is policy- and embodiment-agnostic, allowing us to serve customers with different robots, sensors, policy stacks, and deployment needs.
idoubi@idoubicc · 中文博主 · 12 天前
服务内容标准化 多路开花美滋滋。☺️
idoubi.ai/zh/services
Thomas Wolf@Thom_Wolf · 创始人 · 12 天前
all props to
@stephenwitt
for the analogy