News & analysis, rated.
Breaking AI developments, in-depth guides, real-world case studies, and analysis — each one rated so you know what matters.
Apple overhauls Siri to compete on AI—but execution remains unclear
Apple is investing in Siri improvements to close the gap with ChatGPT and Google Assistant, but details on what will ship, when, and how well it will work remain sparse.
Musk: SpaceX AI satellites will rely on existing tech, not novel systems
Elon Musk said SpaceX's planned AI satellites will use mostly existing technology ahead of the company's expected IPO. Details remain sparse on what 'existing' means and when deployment begins.
OpenAI files for US IPO after Anthropic leads AI startups to public markets
OpenAI has filed for a US initial public offering, following Anthropic's move to go public. The filings mark a watershed moment for AI companies seeking capital and legitimacy.
China's AI exports beat forecast as chip demand surges
Chinese companies are shipping more AI hardware and software than expected, riding global demand for inference chips and tools. What's driving the surge and what it means for Western supply chains.
China allocates $295B for nationwide AI infrastructure push
Beijing is committing $295 billion to build out AI compute capacity and research across the country. The plan signals China's intent to close the gap with the U.S. on model development and chip production.
25 MLOps Guidelines for Model Deployment Now Public
A gray literature review of 103 sources synthesizes state-of-practice MLOps rules for integrating and deploying ML models. Researchers organized findings into five categories to guide architectural decisions.
Deeper transformers need smarter residual routing, not just fixed weights
New method adds directional detail to residual connections in 48-layer transformers, cutting validation loss 4.5% on language modeling tasks without extra parameters.
macOS Agents Fail Where Linux Ones Succeed: New 421-Task Benchmark Reveals the Gap
MacArena benchmarks computer-use agents on 50 macOS apps. Top models drop 26% on native tasks, exposing why GUI agents trained on Linux don't generalize to Apple Silicon.
Deep learning model hits 85% accuracy on polymer sorting with terahertz spectroscopy
Researchers paired terahertz dual-comb spectroscopy with a custom neural network to classify 12 polymer types, including multilayer films and blends. The work suggests a path toward automating plastic recycling quality checks.
FAIR-Calib cuts quantization errors in diffusion LLMs by protecting fragile token decisions
A two-stage calibration method accepted at ICML 2026 reduces frontier decision flips in quantized diffusion language models. New approach targets the stability lag that makes early token commits vulnerable to rounding errors.
Elmes* builds 330 scenarios to test how LLMs teach, not just what they know
Researchers built Edu-330, a benchmark covering 330 educational scenarios across 11 subjects and 3 grade bands, to measure teaching quality rather than factual recall. Top LLMs show stark differences in creativity and scaffolding ability.
Multi-agent framework enriches argument mining with formal structure
CAF-Gen uses a Creator-Reviewer pipeline to automatically transform shallow argument structures into Carneades Argumentation Framework models. The iterative approach mitigates single-pass generation failures.