
Why ByteDance is asking MediaTek to build its AI chips
ByteDance is nearing a custom AI chip deal with MediaTek, blocked from NVIDIA's best hardware by US export rules. Inside AI's custom-silicon era.
Models, tools and the ideas behind intelligent systems.

ByteDance is nearing a custom AI chip deal with MediaTek, blocked from NVIDIA's best hardware by US export rules. Inside AI's custom-silicon era.

Google put four AI chips in orbit on October 1 to test space-based data centers. SpaceX wants a million more. What the first flight is really proving.

ChatGPT, Gemini, Claude, Perplexity and NotebookLM all run free for students — plus the student-only deals worth claiming and the expired promos to ignore.

Meta's 816B-parameter Muse Spark Open is downloadable and runs on two Mac Studios. The specs are real — so are the catches. What the week's open-AI wave means.

NVIDIA unveiled cuPotential at SEMICON West 2026, betting on 'AI-defined fabs.' We unpack what the benchmarks really show — and what they don't.
Distinguish calibration from classification accuracy, read reliability diagrams carefully, and evaluate calibrated models without leaking training information.
Connect vibration measurements, condition indicators and machine-learning outputs without mistaking anomaly detection for a prediction of equipment failure.
Build a clearer evaluation plan for retrieval-augmented AI by separating search quality, supported answers and coverage instead of trusting one overall score.
Explore lower-precision weights, calibration and the evaluation needed before calling a compressed model ready for real use.
How semantic search connects an AI answer to relevant documents—and why finding a source is only the beginning of checking a claim.
Your reading list is empty on this page. Bookmark an article to revisit it later.
A considered collection of technology articles. Explore the latest, or take your curiosity somewhere new.