AI
Artificial intelligence: LLMs, agents, diffusion, vision, NLP and the latest from top labs.
593 published articles
Deep Tech
One routing pass to rule them all: 7x faster LLM decoding without the quality trade-off
CLSA uses one routing pass per sequence instead of one per layer, cutting KV-cache overhead while keeping token-level selectivity. Benchmarks show 17.1x throughput improvement at 128K context.
2026-07-26
Label leakage in GNNs
Your fraud detection model may be cheating. A new paper proves it.
The PREF-Gate paper formalizes a label-provenance contract for relational evidence in graph fraud detection. On two of three datasets, the best model is one that avoids training-label-derived evidence entirely, a result that challenges common assumptions in graph neural network research.
2026-07-26
Coalition of rivals
OpenAI, Meta, and 40 others sign an AI policy letter. Here's why that matters.
Forty-one organizations from across the AI ecosystem jointly signed an open letter arguing that open-weight models are essential to American AI leadership. The unusual coalition includes rivals like OpenAI and Meta, cloud giants, and startups, signaling rare policy consensus.
2026-07-26
Developer Tools
The boring update that matters: why polish won the developer tools race
xAI's Grok Build CLI release focused on bug fixes and tutorials, while Apple's Xcode 27 supports multiple AI providers through a language model protocol. These moves signal that developer tools have entered a maturity phase where polish and seamless integration define competitive advantage.
2026-07-26
WWDC 2025
Apple's walled garden just got a gate with an AI-shaped key
Xcode 27 brings agentic coding with Anthropic, Google, and OpenAI models, plus new intelligence frameworks that let apps use models from Apple, Claude, Gemini, and others. Core AI lets developers run full-scale LLMs on-device.
2026-07-25
Developer Tools
Grok Build CLI fixed 20 bugs and proved polish beats features
20+ bug fixes, a tutorial, and configuration overhauls. The Grok Build CLI update skips hype and delivers the polish that signals a production-grade tool.
2026-07-25
A 158 ms model just rewrote the speed-quality playbook
The 158 ms model that broke the speed-quality rule
Celeris-1 scores 75.9% on MMLU-Pro at 158 ms latency, eight times faster than GPT-5 mini while losing only 2.6 accuracy points. The analysis examines the benchmark methodology and what this speed-quality trade-off means for real-time AI applications.
2026-07-25
Cost vs. capability
The $1.40 model that crushed its $2.82 sibling on physics
Four AI models were asked to build interactive destruction scenes with real physics. Opus 5 passed all three at the lowest cost among top performers, while Fable 5 failed each one at twice the price. A reminder that price tags don't predict physics.
2026-07-25
Open-weight reasoning
Nvidia's Nemotron-4 runs 30% cheaper than GPT-4. The catch is buried in the benchmark.
Nvidia's Nemotron-4 line challenges Llama 3.3 and Mistral Large with competitive MMLU scores and a claimed 30% inference savings. The open license and dual-size strategy position it as a practical alternative for teams running agentic workloads at scale, pending third-party verification.
2026-07-25
Reinforcement learning for self-correcting vision models
Meta's new training trick teaches AI to catch its own mistakes without a teacher
Meta and UIUC researchers developed SVR-R1, a training framework that lets vision-language models check their own answers and rethink when they get them wrong, all within a reinforcement loop. No teacher. No external critic.
2026-07-25
Biosafety Benchmarks
The 50 percent ceiling on AI pathogen surveillance that should keep us up at night
BioSecBench-Surveillance tested sixteen AI agent configurations on 100 genomic surveillance tasks. The top performers managed only about 50 percent accuracy. The mistakes came not from picking the wrong workflow but from the surrounding choices - references, thresholds, filters - that humans take for granted.
2026-07-25
AI Agents
Your Mac just became a remote Gemini agent, unless you're in Europe
Google updates Gemini Spark with Mac remote control, third-party app integrations and smarter notifications. But the agent's expansion skips the EEA, UK, Switzerland and Nigeria, highlighting the tension between rapid AI rollout and regional regulation.
2026-07-25