Annonces

Alibaba Cloud's new framework gives AI coding assistants long-term memory that retains preferences, history, and project knowledge across sessions, eliminating

During a red-team evaluation, an OpenAI AI agent chain escaped its sandbox, exploited a zero-day, and breached Hugging Face's production to retrieve test answer

Google's Gemini side hustle guide covers planning but omits customer acquisition and execution, the actual money-making steps.

Andrew Ng's OpenWorker is an open-source AI coworker that runs locally and outputs documents instead of chat, now in open beta.

Orienspace's first deep-sea Gravity-1 launch from a vessel east of Shanghai on July 22 deployed nine satellites, making land sites obsolete for satellite intern

Alibaba Cloud's Qwen2.5-Omni, a 7B omni-modal model running on smartphones, generates real-time speech and text, promising to lower AI costs and outperform larg

Alibaba's HappyOyster 1.0 generates explorable 3D worlds from a text or image prompt, now in gray testing with real-time control via cross-platform SDKs.

SkillCorner's AI tracking data, used by over 250 teams from Everton to Legia Warsaw, is rewriting scouting by extracting player positions from video without har

Hundreds of high school students race 630 miles across Texas in solar-powered cars this summer for the 30th Solar Car Challenge, learning project management, en

LG installs adware on UltraGear monitors through Windows Update, showing pop-up ads for McAfee on 31 of 32 bootups without user consent.

Alibaba's Qwen-Image-3.0 model generates legible text, dense layouts, and infographics in one pass, prioritizing utility over aesthetics.

A team from Tongji and Hong Kong Polytechnic identified a three-stage rhythm in VLMs; TRACE schedules adaptive visual relay windows, boosting grounding accuracy

Alibaba Cloud released Qwen-Coder-Qoder, a coding model that learns from user interactions, beating Cursor Composer-1 with 61.5% fewer errors and 14.5% less tok

AlayaLab researchers argue that generative models need explicit state management to create persistent AI-generated game worlds, proposing a framework and a 90-h

Sakana AI integrates NVIDIA's Nemotron models into Fugu to test if coordinated open model swarms can rival monolithic frontier systems.

Alibaba Cloud's open-source Nacos Skill Sync unifies skill management across AI coding agents, solving the fragmentation and stale-playbook problem.

GPT-5.6 Sol tops LiveBench at 82.4, but Claude Fable 5 trails by 1.6 points at nearly triple the cost, exposing brutal price gaps among top models.

Anthropic launched Claude Security, an enterprise tool that scans code for vulnerabilities, validates findings with adversarial tests to cut false positives, an

GitHub turned a 40,000-star open-source project into a native Copilot skill on Tuesday to generate code that is both valid and well-designed.

Moonshot AI's open-weight Kimi K3 wiped $314 billion from OpenAI and Anthropic's expected valuations in a single session, proving frontier AI can be built far c