1846 articles
Ant Group released LLaDA2.0, a framework for scaling diffusion language models up to 100 billion parameters. The method converts existing auto-regressive models while boosting efficiency and performance.
Benchmark results show DeepWriter hitting the highest HLE score among leading agentic AI tools. The data reveals significant differences in long-form reasoning and autonomous performance capabilities.
Nvidia launched Nemotron-Cascade, a reinforcement learning framework designed specifically for reasoning-focused AI models. Testing shows consistent accuracy improvements through staged training phases.
Fresh benchmark data reveals Claude Opus 4.5 leading the pack in long-duration AI task performance, sustaining successful execution for nearly five hours at 50% success probability.
OpenAI's profit margins on compute costs for paying customers have surged to nearly 70% by October, driven by cheaper infrastructure and smarter inference technology.
Fresh benchmark data reveals Claude Opus 4.5 holds up better than GPT-5.1-Codex-Max when tasks stretch into multi-hour territory, with the performance gap widening as duration increases.
A new benchmark result puts GPT-5.2 in the spotlight, while version tracking reveals how far apart major AI platforms have drifted in their development cycles.
Robotics startup TARS has showcased an autonomous embroidery robot and demonstrated how the same capability can be applied to industrial wire harness assembly.
Takeda reports positive Phase 3 results for zasocitinib, an AI-assisted oral TYK2 inhibitor for psoriasis, with 50% of patients achieving near-complete skin clearance by Week 16.
LimX Dynamics is using AI to revolutionize biped robot design, creating fully modular systems that optimize balance and movement through data-driven engineering.
A debate over Oracle's AI strategy highlights growing model commoditization and questions whether the market is heading toward winner-take-all outcomes or staying fragmented.
Google has signed a long-term power agreement with NextEra Energy to restart Iowa's Duane Arnold nuclear plant, with electricity expected to begin flowing by 2029.
Grok.com saw steady month-over-month traffic growth throughout 2025, climbing from 1.2 million visits in January to over 227 million by November.
A massive AI facility running 100,000 GPUs at full tilt is exposing hard limits in power delivery, cooling systems, and materials engineering. The real-world data shows that physical infrastructure—not just chip performance—is now the defining constraint for AI expansion.
An early AGI forecast from 2009 projected human-level artificial intelligence around 2028. A probability chart illustrates how expectations were modeled long before recent AI advances.
METR evaluation data shows Claude Opus 4.5 achieving a long median task horizon, while maintaining a much shorter high-reliability threshold.
Epoch AI's latest benchmark shows Gemini 3 Flash keeping pace with top models on easier FrontierMath tiers but struggling on the toughest problems.
MiniMax-M2.1 demonstrates superior multi-agent coordination compared to its predecessor, successfully managing five parallel subagents in Claude Code CLI while MiniMax-M2 encounters critical tool execution errors.
New research reveals SonicMoE method delivers major efficiency improvements for AI models on Nvidia H100 GPUs, boosting speed while cutting memory usage nearly in half.
New survey reveals majority of U.S. voters now use AI services at least once per week, with ChatGPT dominating the market
We use cookies to improve your experience on our site and to show you relevant advertising. To find our more, read our privacy policy and cookie policy
Artem Voloskovets
Peter Smith
Saad Ullah
Marina Lyubimova
Eseandre Mordi
Sergey Diakov
Alex Dudov
Usman Salis
Victoria Bazir
