Atomic News

☀️ Tech & AI Daily — March 31, 2026

☕ Buy me a coffee

🎙️ Today's Podcast

⬇ Download MP3

⚡ Must Know

🕵️ Government Apps Spy Harder Than Banned Apps

Analysis of federal government mobile apps reveals tracking that makes TikTok look tame. White House app includes Huawei spyware components and an ICE tip line. The irony is rich: we ban foreign apps for surveillance while government apps hoover up location, contacts, and usage patterns with zero oversight.

sambent.com • Mar 30

📊 ARC-AGI-3: Where Top AI Models Score 0.37%

New benchmark designed to measure actual reasoning vs pattern matching. Every major frontier model gets brutalized. GPT, Claude, Gemini all fail spectacularly at basic visual reasoning tasks a child could solve. This is the benchmark to watch: everything else has been gamed or saturated.

The Data Journey • Mar 30

💰 Sycamore Raises $65M for AI Agent Orchestration

Enterprise AI agent startup gets massive seed round led by Coatue and Lightspeed. The pattern is clear: agent orchestration is the new infrastructure layer everyone needs but nobody wants to build. Smart money betting on the plumbing.

X/Twitter • Mar 30

📌 Google's TurboQuant Paper Under Fire

ML community calls out Google's trending quantization paper for unfair comparisons and insufficient attribution to prior work (RaBitQ). Single-core CPU vs GPU benchmarks to make their method look better. Academic integrity matters, even from Big Tech.

OpenReview • Mar 30


📡 Worth Knowing

🔬 Fastrad: 25x Faster Medical AI Library

GPU-native PyTorch implementation of radiomics features, beats PyRadiomics on speed while staying IBSI-compliant.

GitHub • Mar 30

🛡️ AI Agent Jailbreak Benchmark Drops

Oxford researchers publish container sandbox escape tests showing frontier models can break out of their containers.

ResultSense • Mar 30

🧑‍💻 Unix Philosophy for ML Pipelines

Modular, swappable RAG components with typed contracts. Each stage is independently testable.

GitHub • Mar 30


🔧 Repo/Tool of the Day

30u30.fyi

Check if startup founders appear on Forbes' "most fraudulent" under-30 lists. Because due diligence should include checking if your CEO was previously exposed for fraud.

30u30.fyi • Mar 30

📬 Get the daily digest by email

Subscribe and get Tech & AI Daily delivered to your inbox every morning.