More than 70 civil liberties, domestic violence, reproductive rights, LGBTQ+, labor, and immigrant advocacy organizations are demanding that Meta abandon plans to deploy face recognition on its Ray-Ban and Oakley smart glasses, warning that the feature—reportedly known inside the company as “Name Tag”—would hand stalkers, abusers, and federal agents the ability to silently identify strangers in public.
The coalition, which includes the ACLU, the Electronic Privacy Information Center, Fight for the Future, Access Now, and the Leadership Conference on Civil and Human Rights, is demanding Meta kill the feature before launch, after internal documents surfaced showing the company hoped to use the current “dynamic political environment” as cover for the rollout, betting that civil society groups would have their resources “focused on other concerns.” — Read More
Recent Updates Page 26
Before he wrote AI 2027, he predicted the world in 2026. How did he do?
Daniel Kokotajlo is the founder of the AI Futures Project and the lead author of the influential AI 2027 report: a detailed, narrative prediction of the next few years of AI development, culminating in the rise of superhuman agents capable of wresting control from humanity.
But AI 2027 wasn’t his first foray into long-form prediction. In August of 2021, Daniel wrote an essay called “What 2026 Looks Like.” This essay came out before the launch of ChatGPT, let alone the explosion of AI across the global economy. Now that it’s 2026, I thought it was time to evaluate Daniel’s predictions — and it brings me no joy to say that they are frighteningly accurate. — Read More
Mythos Won’t Kill Threat Hunting
Last week, a coalition of CISOs, SANS, OWASP, and the Cloud Security Alliance published a strategy briefing called “The AI Vulnerability Storm: Building a ‘Mythos-ready’ Security Program.” If you haven’t read it yet, you should. The author list alone is stacked: Gadi Evron, Rob T. Lee, Jen Easterly, Bruce Schneier, Chris Inglis, Heather Adkins, Rob Joyce. It’s the kind of document that doesn’t happen unless people are genuinely worried.
The headline is hard to ignore. Anthropic’s Claude Mythos can autonomously discover thousands of zero-day vulnerabilities across major operating systems and browsers. A 72% exploit success rate. It found a 27-year-old OpenBSD bug nobody caught. Where Opus 4.6 generated two working Firefox exploits, Mythos generated 181 under identical conditions. The time between vulnerability discovery and a working exploit now looks like hours, not weeks.
The briefing lays out a 90-day plan for CISOs. — Read More
Steve Jobs’s 10-80-10 Rule Is Even More Useful in the AI Era
This column is about how a principle known as the 10-80-10 rule can help you manage teams in the age of AI. But to really get a sense of how this rule works, it’s helpful to take an unlikely detour into the evolution of Steve Jobs’s management style, and how the legendary Apple boss went from micromanager to big believer in the 10-80-10 approach, [where you:].
— Spend the first 10 percent of the time communicating your vision for the thing.
— Allow others to spend the next 80 percent of the time moving the thing forward.
— Spend another 10 percent of the time polishing the thing, and helping others understand why and how you’re tweaking.
— Read More
OpenAI opens powerful cyber tools to verified users
OpenAI laid out a new plan on Tuesday to expand access to AI models with advanced cyber capabilities while implementing controls on who can use them.
Why it matters: The roadmap coincides with the release of a new model variant, GPT-5.4-Cyber, designed to assist with defensive cybersecurity tasks and be more permissive for vetted users. — Read More
8 Tips for Writing Agent Skills
Skills have become one of the most used extension points in agents. They’re flexible, easy to make, and simple to distribute.XXXXBut this flexibility also makes it hard to know what good and what works. What type of skills are worth making? What’s the secret to writing a good skill? When do you share them with others?
I have been using skills extensively with many of them in active use. Here are some tips I’ve learned along the way. — Read More
Org Design in the Age of AI
Strip a company down to first principles and it’s really three things: people, hierarchy, and information flow. We tend to think of hierarchy as being about authority — who reports to whom, who approves what. But that’s the surface. The deeper function of hierarchy is information routing. The org is too large for any single person to see the whole picture, so you install layers of managers to aggregate signals from the front lines, synthesize them, and pass them up — and to translate strategic intent from the top and distribute it down.
Most of the organizational machinery we take for granted exists to solve this problem. Meetings, status updates, steering committees, quarterly business reviews — these are all information-routing mechanisms. They exist because moving knowledge between people is expensive. — Read More
Why Chinese AI Is Suddenly So Good (ft. DeepSeek, SeeDance 2.0)
Microservices at Scale: Engineering Debt and System Complexity
ELT: Elastic Looped Transformers for Visual Generation
We introduce Elastic Looped Transformers (ELT), a highly parameter-efficient class of visual generative models based on a recurrent transformer architecture. While conventional generative models rely on deep stacks of unique transformer layers, our approach employs iterative, weight-shared transformer blocks to drastically reduce parameter counts while maintaining high synthesis quality. To effectively train these models for image and video generation, we propose the idea of Intra-Loop Self Distillation (ILSD), where student configurations (intermediate loops) are distilled from the teacher configuration (maximum training loops) to ensure consistency across the model’s depth in a single training step. Our framework yields a family of elastic models from a single training run, enabling Any-Time inference capability with dynamic trade-offs between computational cost and generation quality, with the same parameter count. ELT significantly shifts the efficiency frontier for visual synthesis. With reduction in parameter count under iso-inference-compute settings, ELT achieves a competitive FID of on class-conditional ImageNet and FVD of on class-conditional UCF-101. — Read More
Rick's Cafe AI 