Tech news, summarized and translated.
The European Commission hits Google with a record fine under the Digital Markets Act for favoring its own services in search results and restricting app developer steering.
Commodore's Amiga 1000 launched 40 years ago with preemptive multitasking and custom graphics/sound hardware that the rest of the industry wouldn't match for a decade. A look at what made it special — and why it failed commercially.
New DNA analysis of Francesco I de' Medici's remains confirms malaria, not arsenic poisoning, as the cause of his and his wife's deaths in 1587, ending a 400-year-old murder mystery.
Cactus Hybrid post-trains small models to output a confidence score with every answer, enabling reliable cloud handoff only when needed. The first release, Gemma 4 E2B Hybrid, routes just 15–35% of queries to a larger model while matching Gemini 3.1 Flash-Lite on benchmarks.
Mitchell Hashimoto argues that SIMD is simpler than its reputation suggests, offering a five-step pattern that turns naive loops into vectorized code with dramatic speedups.
A deep dive into the Perlin noise algorithm, from the fundamentals of coherent noise to the implementation details of gradient vectors and permutation tables.
Tens of billions in AI infrastructure debt is collateralized by GPU clusters whose value depends on operational state—and lenders have no way to measure that risk.
A federal judge approved a $1.5 billion settlement where Anthropic will pay authors ~$3,000 per book for using pirated copies to train Claude. The ruling upholds fair use for training but punishes the source of the data.
Eclipse Vert.x brings reactive, non-blocking I/O to IoT with MQTT, AMQP, and Kafka support — all on the JVM.
A personal reflection on happiness from the late Pieter Hintjens, creator of ZeroMQ.
The latest Vert.x release brings HTTP/2 support, a pluggable service discovery system, a circuit breaker implementation, and a bridge to Apache Camel for enterprise integration.
A head-to-head test of four frontier AI models using a simulated colored-pencil toolset reveals GPT-5.6 Sol as the clear winner on both cost and quality, while Claude Fable 5's high cost and verbose tool use hurt its performance.