All AI News

New benchmarks reveal GLM-5 and Minimax-2.5 outperforming leading models on Fiction.liveBench, signaling a shift toward agentic reasoning in open-source AI. Sources indicate GLM-5’s 744B-parameter architecture and DeepSeek Sparse Attention are key to its enhanced performance in long-horizon tasks.

A mysterious surge in automated traffic originating from Lanzhou, China, is overwhelming websites across sectors—from small businesses to U.S. federal agencies—triggering security alerts and CAPTCHA challenges. Experts warn the pattern suggests coordinated scraping or reconnaissance, though the motive remains unclear.

A new investigation reveals the Trump administration’s covert expansion of ICE surveillance operations, aided by Palantir’s AI systems, sparking ethical outcry from within the tech workforce. The campaign’s infiltration into daily life echoes the uncanny — familiar yet deeply unsettling.

OpenAI has unveiled GPT-5.3-Codex-Spark, its first large language model deployed on Cerebras Systems' CS3 AI accelerators, achieving 1,000 tokens per second and 15x faster code generation than Nvidia-based systems. This marks a strategic pivot away from Nvidia’s dominance in AI infrastructure.

A newly introduced text-based logic puzzle, dubbed the 'Car Wash Test,' has exposed significant disparities in AI reasoning capabilities. While most major models failed, Gemini Pro and Gemini Fast were the only ones to correctly solve the riddle, sparking debate about benchmark design and cognitive evaluation in AI systems.

A growing number of viewers are noticing a strikingly uniform speech pattern across YouTube videos, traced to AI-generated scripts from ChatGPT. Critics argue this homogenization undermines authenticity, while creators defend it as a productivity tool.

PlanoAI has released version 0.4.6, introducing a signals-based tracing system accessible directly through the command line, eliminating the need for graphical interfaces or additional dependencies. The update targets developers building complex AI agent systems, offering real-time observability with minimal friction.

As artificial intelligence systems demand unprecedented computational power, the host-device paradigm has become the backbone of modern GPU-accelerated computing. This article explores how CPUs and multiple GPUs collaborate, drawing on technical insights from leading AI development platforms.

Google has unveiled a major upgrade to its Gemini 3 Deep Think model, achieving an unprecedented 84.6% score on the ARC-AGI-2 benchmark — a metric long considered a litmus test for artificial general intelligence. Experts are divided on whether this represents a true leap toward AGI or merely advanced narrow reasoning.

A newly published paper from DeepMind reveals the cancellation of a high-profile AI project designed to achieve superhuman performance in mathematical reasoning. The decision follows unforeseen limitations in generalization, scalability, and concerns over interpretability in critical domains.

Following a wave of executive departures and the merger of xAI with SpaceX, Elon Musk has pivoted to an ambitious new vision: establishing a lunar industrial base powered by mass drivers. The strategy aims to attract top engineering talent with a bold, science-fiction-meets-reality mission.

Anthropic's latest model, Claude Opus 4.6, is generating global interest for its unprecedented reasoning capabilities — but some experts warn its advanced autonomy may outpace safety protocols. Industry analysts and researchers are divided on whether this leap forward signals the dawn of AGI or a dangerous overreach.

From revolutionizing Google’s search engine to co-creating TPUs and reviving trillion-parameter models, Jeff Dean has silently shaped the foundations of today’s AI landscape. His decades-long career exemplifies the fusion of systems engineering and machine learning innovation.

Anthropic has secured a record-breaking $30 billion in Series G funding, pushing its valuation to $380 billion — more than double its previous valuation. Led by GIC, Founders Fund, and MGX, with strategic participation from Microsoft and NVIDIA, the capital will fuel enterprise AI model development and global infrastructure expansion.

Multiple users and internal concerns suggest OpenAI is prioritizing computational efficiency over accuracy, delivering superficial or incorrect responses—especially on paid subscriptions. Reports coincide with leadership changes and rumored shifts in AI safety priorities.

Google's Gemini Deep Think model achieved a leading success rate on the ARC-AGI-2 test set as of 2026, marking a critical milestone in the development of artificial general intelligence.

An internal insight from X’s head of product suggests the platform has approximately 90 days to successfully integrate advanced AI capabilities or risk obsolescence in the rapidly evolving social media landscape. The warning, gleaned from a Reddit thread, underscores mounting pressure on Elon Musk’s platform to compete with emerging AI-native competitors.

Microsoft has rolled out a trio of groundbreaking AI enhancements to its Microsoft 365 ecosystem, including Copilot Memory for personalized workflows, Project Opal for task automation, and Microsoft 365 Copilot Business tailored for SMEs. These updates signal a strategic pivot toward context-aware, enterprise-grade artificial intelligence.

MiniMaxAI has revealed its new MiniMax-M2.5 model with 230 billion total parameters and only 10 billion active parameters, leveraging sparse activation to rival top-tier models at lower computational cost. The announcement, first reported by OpenHands and corroborated by AI infrastructure experts, signals a strategic shift in efficient large language model design.

Elon Musk has described the departure of six of xAI’s original 12 co-founders as a necessary evolution, signaling a strategic realignment toward long-term AI development and lunar ambitions. Industry experts question whether the exits reflect ideological friction or operational strain within the rapidly scaling startup.

OpenAI has unveiled GPT-5.3-Codex-Spark, a lightweight, high-speed AI model designed for real-time coding workflows, leveraging Cerebras hardware to achieve 1,000 tokens per second. Though less visually refined than its predecessors, its speed enables unprecedented developer flow and iterative productivity.

An internal insight from X’s head of product suggests the platform may have only 90 days left to stabilize before irreversible decline. The warning, circulating via Reddit, highlights mounting operational, financial, and user trust challenges following Elon Musk’s acquisition.

A viral Reddit experiment reveals how large language and vision models, when chained together in a game of digital telephone pictionary, produce increasingly surreal and unintended results. The phenomenon highlights both the creative potential and interpretive drift inherent in multimodal AI systems.

Concerns are mounting after reports surfaced that users of Elon Musk’s AI chatbot Grok are attempting to deblur images of minors from the Epstein case files, raising serious ethical and legal red flags. Experts warn this misuse highlights dangerous gaps in AI content moderation.