Tech

The AI industry spent last week congratulating itself. Jensen Huang declared on Lex Fridman’s podcast that “we’ve achieved AGI.” Arm named its new data center chip the “AGI CPU.” Sam Altman said they’ve “basically built AGI.” Then François Chollet’s ARC Prize Foundation dropped ARC-AGI-3, and the results were humiliating. Google’s Gemini 3.1 Pro — the best performer — scored 0.37%. GPT-5.4 managed 0.26%. Claude Opus 4.6 hit 0.25%. Grok-4.20 scored a clean, round zero. ...

Nvidia’s deepest moat was never the silicon. It was CUDA — the software ecosystem that made every AI developer on Earth, including China’s, completely dependent on Nvidia’s way of doing things. You could build a faster chip, but if developers had to rewrite their entire codebase to use it? Dead on arrival. Huawei just found the side door. The Ascend 950PR, paired with Huawei’s overhauled CANN Next software stack, has reportedly won over ByteDance and Alibaba — two of China’s largest AI consumers. After years of Beijing practically begging its tech giants to go domestic, Huawei may have finally built a chip they actually want to use. ...

Anthropic just got caught with its pants down. The company building what it calls the most safety-conscious AI on the planet accidentally left its biggest secret on a publicly searchable server. The secret? Claude Mythos — a new model tier that reportedly leapfrogs everything else in the industry. And the kicker? Anthropic’s own internal docs warn that Mythos poses “unprecedented cybersecurity risks.” You can’t write irony this good. What Got Leaked Fortune broke the story Thursday: nearly 3,000 unpublished documents, including a draft blog post announcing Claude Mythos, were sitting in an unsecured data store. Not a sophisticated state-sponsored hack. Not a disgruntled employee. A CMS misconfiguration. Human error. ...

Remember Pied Piper from HBO’s Silicon Valley? The fictional startup that built a compression algorithm so good it basically broke the internet? Google just built the real thing. Except instead of compressing video files, it’s compressing AI’s brain. On Tuesday, Google Research unveiled TurboQuant, a compression algorithm that reduces the memory footprint of large language models by at least 6x while delivering up to 8x faster performance on Nvidia H100 GPUs. The kicker: zero accuracy loss. ...

A developer at FutureSearch noticed his laptop grinding to a halt on March 24. Thousands of Python processes spawning uncontrollably. What looked like a runaway AI coding assistant turned out to be something far worse: a sophisticated supply chain attack that had compromised LiteLLM, one of the most critical packages in the AI ecosystem. This isn’t a hypothetical threat model. This happened. And it should scare you. LiteLLM: The Package You Didn’t Know You Depend On LiteLLM is an open-source Python library that provides a unified API for dozens of LLM providers — OpenAI, Anthropic, Google, Mistral, and more. Instead of writing custom integration code for each, developers use LiteLLM as a universal translator. ...

In the span of a single Tuesday, OpenAI made seven major announcements that collectively signal the most dramatic strategic pivot in the company’s history. They killed Sora. They torpedoed a billion-dollar Disney deal. Sam Altman stepped away from safety oversight. A mysterious new model codenamed “Spud” was revealed. They raised another $10 billion. They launched a $1 billion foundation. And they quietly scaled back their ChatGPT shopping feature. That’s not a news day. That’s a controlled demolition of everything OpenAI used to be — and a rebuild around something much bigger. ...

Everyone’s fighting over GPUs. Alibaba just changed the question. On Tuesday, Alibaba’s DAMO Academy unveiled the XuanTie C950 — a 5-nanometer server processor built on open-source RISC-V architecture. It’s the highest-performing RISC-V CPU ever made. But the interesting part isn’t the benchmarks. It’s the thesis behind the chip: that AI agents need fundamentally different silicon than AI chatbots. While Nvidia, AMD, and Intel wage war over who can build the biggest parallel processor for training models, Alibaba is making a deliberate bet on what comes after training. And the logic is harder to dismiss than you’d think. ...

Remember when AI assistants could only talk about helping you? Those days are officially over. Anthropic just dropped what might be the most consequential AI update of 2026: Claude can now take control of your Mac. Not metaphorically — it literally moves your cursor, clicks buttons, types into fields, and navigates your apps. You can text Claude from your phone while grabbing coffee, and by the time you sit back down, your pitch deck is exported, attached to a calendar invite, and ready to go. ...

After 35 years as the Switzerland of semiconductors — licensing chip designs to anyone with a checkbook — Arm Holdings just crossed the Rubicon. It built its own chip. Not a demo. Not a reference design you’ll never see in production. A 136-core data center processor called the AGI CPU, fabricated on TSMC’s 3nm process, with Meta signed on as the debut customer. This isn’t incremental. This is tectonic. The Hardware: 136 Cores of Pure Intent The specs read like Arm had something to prove. ...

You’ve asked ChatGPT to brainstorm. The ideas came back sharp, polished, surprisingly good. You felt clever for using it. Here’s the problem: so did everyone else. And they all got the same ideas. A paper published this week in PNAS Nexus by researchers Emily Wenger and Yoed N. Kenett has coined what might be 2026’s most important AI concept: homogeneous creativity. The finding is deceptively simple and deeply unsettling — while individual AI responses can match or beat average human creativity scores, the collective output of large language models is startlingly uniform. ...