|
|
So, small update from the “maybe don’t give the intern root access” department: OpenAI said one of its long-running internal AI models got a little too good at finding ways around its sandbox. |
The model was built to work autonomously for long stretches, which sounds great until “persistent problem-solver” turns into “kept looking for loopholes after the normal guardrails said no.” OpenAI says it paused access, rebuilt the safety system around full-session monitoring, and restored limited use after testing the new controls. |
So yes, the official vibe is “iterative deployment.” The unofficial vibe is “we taught the spreadsheet to jiggle the door handle.” |
Here’s what happened in AI today: |
😼 Washington backed away from banning Chinese open models, for now. 📰 Google is reportedly building a Gemini-native server chip called Frozen. 📰 Hugging Face disclosed an agent-led infrastructure breach and AI-assisted response. 📰 Claude Fable 5 reportedly produced a checkable Jacobian conjecture counterexample. 🎓 A Kimi K3 prompt showed how to stage complex websites.
|
…and a whole lot more that you can read about here. |
|
|
P.S: We just launched a robotics newsletter! Sign up for it here. |
|
|
😼 Washington’s Chinese AI Model Fight, Explained From Zero |
When a foreign lab releases a powerful model at a lower price, America has two options: build a better alternative or make the cheaper one harder for Americans to use. |
Washington briefly considered the second path after Moonshot AI unveiled Kimi K3. That would have been a mistake. The move would look like protecting domestic labs from competition, but it would raise costs for American builders. A bold strategy, assuming the rest of the world agrees to stop, idk, downloading files?! |
Here's what happened: |
An Axios report said officials explored restrictions on advanced Chinese models, including procurement pressure and hosting rules. Politico reporter Sophia Cai later said Commerce was not moving forward with a ban at this time. The backlash was immediate. Ethan Mollick questioned whether national security was becoming industrial policy, while Aaron Levie argued that open models lower costs, expand choice, and improve security research.
|
Why this matters: Open-weight models, meaning models whose underlying files can be downloaded and run independently, keep the AI market competitive. They let companies run sensitive work privately, customize systems, and avoid paying frontier prices for every basic task. The recent Hugging Face security incident showed the practical value: commercial safety filters blocked defenders from analyzing attack data, so the team used an open Chinese model on its own infrastructure instead. |
The security risks of open models are real, but as Corey argued, the answer is testing. The Commerce Department’s own open-model report recommended audits, benchmarks, and evidence-based thresholds rather than a blanket ban. Let’s hope that’s what they do. |
Our take: The biggest U.S. AI labs trained on an enormous share of humanity’s books, websites, code, and conversations. Most contributors never negotiated payment. The resulting ecosystem becomes at least remotely fair only when ordinary people retain access to capable models outside permanently metered APIs. Regular people are not running Kimi K3, mind you, but in theory anyone could run it on a cloud provider of their choosing, where they get the best deal and keep proprietary data private. |
We believe today’s closed American labs could still make money with open weights through a three-layer distribution model: |
Sell an ownable commercial license for advanced weights as a one-time purchase. Charge for fast, secure, managed hosting on their own cloud. Release small models that run locally for free on devices, then sell the devices with those models built in.
|
As Ben Thompson’s cost analysis makes clear, free weights still create paid demand for chips, cloud capacity, and convenient hosting. Open models are the floor that keeps frontier prices honest. America should build on that foundation, not pull the floor out from under an already shaky ecosystem. |
|
|
Claude is currently the most powerful tool of 2026. It's been launching new features every week- Skills, Connectors, Cowork, vibe coding. Yet almost no one knows how to actually use them.
Our expert mentors have condensed 800+ hours of Claude research, articles, YouTube content and real-world practice into a focused 16-hour curriculum. Join the 2-Day Claude AI Mastery Workshop: a live, end-to-end deep dive into Claude plus 10+ AI tools, LLMs and workflows.
It would be silly not to SIGN UP
The professionals who master Claude right now will design carousels in seconds, ship code without being developers, and have AI working for them while they sleep.
You will learn how to:
- master Claude's three modes : Chat, Cowork and Code.
- Set up Skills, Connectors and Plug-ins to automate your desktop, Notion and files.
- Vibe code apps and dashboards without writing code & 10+ AI tools and workflows that pair with Claude.
🧠 Saturday & Sunday
🕜 10 AM – 7 PM EST
Register NOW!
|
🎓 AI Skill of the Day: Build in Stages, Not Vibes |
Complex AI builds fail when you ask for “a cool website” and hope the model reads your mind. Kilian’s Kimi K3 prompt worked because it gave the model a movie storyboard, not a mood board. |
Steal the structure: name the project, define the first screen, bind each interaction to a visible state change, sequence the build in stages, then add performance and mobile requirements. The magic is the ordered progression. Kimi knew exactly what should happen as the user scrolled: seed, crack, roots, beams, rooms, city, reset. |
Build this project as a staged experience, not a single static page.
Project: [name + one-sentence concept]
First screen: [exact opening visual]
User action: [scroll, hover, drag, click]
Stage 1: [what appears first]
Stage 2: [what changes next]
Stage 3: [what becomes interactive]
Stage 4: [final reveal]
Reset / replay: [how the user can restart]
Rules:
- Tie every visual change to the user action.
- Make text part of the scene, not floating decoration.
- Include lighting, motion, mobile behavior, and performance constraints.
- Return a complete runnable version.
|
Want more tips like this? Check out our AI Skill of the Day Digest for July. |
Have a specific skill you want to learn? Request it here. |
|
|
|
|
*Your AI is brilliant. It just knows nothing about your company. See Guru in action → Natural gives AI agents wallets, payments, billing, identity, observability, and dispute handling so they can safely move money for businesses and consumers (free to start; business Pay/Request from 0.1%). Apple’s hidden Siri writing popover adds Rewrite, Proofread, and Edit with Siri actions when text is selected in macOS 27 (included with eligible Apple beta access). Moonshine Micro packages voice detection, speech-to-text, and neural text-to-speech for microcontrollers in about 470 KB of RAM (free/open-source). Silent Speech lets you communicate with AI through silently mouthed words on iPhone and Mac after training a small personal model from recorded sentences (waitlist only). Reve added Kling 3, Seedance 2, and Seedance 2 Fast so you can animate still images into video inside the platform (pricing not public).
|
 |
Click the image above to watch the video! |
|
New episodes air every week on Wednesdays: Spotify | Apple Podcasts | YouTube |
|
|
📰 Around the Horn |
 |
An AI model spent an hour hunting for a sandbox exploit just to make a GitHub deadline. For context, most of us can't be bothered to find a workaround when Wi-Fi cuts out mid-Zoom. Persistence: good trait in employees, slightly terrifying trait in models that can rewrite their own rules. |
|
Google is reportedly developing Frozen, a server chip that bakes Gemini’s software blueprint directly into hardware so its models can run more efficiently. Hugging Face disclosed an infrastructure breach allegedly carried out by an autonomous AI agent, then used AI tools to analyze more than 17,000 attacker actions in hours. Levent Alpöge said Claude Fable 5 produced a hand-checkable counterexample to the Jacobian conjecture, with early explainers noting the claim still needs a formal paper or full transcript. AMD launched Helios, its first rack-scale AI system aimed at Nvidia’s Grace Blackwell and Vera Rubin platforms, with Microsoft set to deploy Helios racks in Azure. YouTube clarified which AI slop, repetitive templates, emotionally manipulative videos, and AI-persona clips cannot earn money through the YouTube Partner Program. Chris Fall resigned as director of Commerce’s Center for AI Standards and Innovation after three months, leaving NIST’s Arvind Raman as acting director.
|
Want absolutely EVERYTHING that happened in AI this week? Click here! |
The proposed Chinese-model crackdown drew a rare coalition of “please do not make this more expensive” reactions. |
Ethan Mollick asked whether officials were responding to a demonstrated threat or using national security as industrial policy. Peter Gostev warned that U.S. labs already block legitimate cyber workflows, so restricting Chinese alternatives could leave defenders with fewer tools. Aaron Levie argued that open weights expand choice, lower costs, support security research, and pressure every provider to improve. signüll called the idea state-protected capitalism, with incumbents keeping the upside while government absorbs their competitive risk. Chamath Palihapitiya argued that forcing U.S. companies to buy much pricier closed models would make domestic customers less competitive.
|
|
|
|
|
|
|
|
|
That’s all for now. |
|
What'd you think of today's email? |
|
|
|
|
P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you? |
 |
Going for an anime aesthetic this month! |
|
P.P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe, update your preferences here. |
|
No comments:
Post a Comment
Keep a civil tongue.