Send us a text 🌍 INAI • The Open AI Hub The Intelligence Atlas → the world’s most comprehensive, open hub of AI knowledge. 2 Million+ tools, models, agents, tutorials & daily news—free for all, updated every day. https://github.com/inai-sandy/inAI-wiki Top Highlights: OpenAI launched a cheaper GPT-5 web search API with domain filtering, Google announced a gigawatt-scale AI hub in Visakhapatnam making it India's first "AI city", OpenAI partnered with Broadcom on custom AI chips and 10 GW infrastructure, Walmart rolled out AI-powered checkout and ChatGPT shopping nationwide, and Alibaba's Qwen3-VL expanded while Veo 3 and Sora 2 competed in AI video. New Tools: OpenAI Search API enables safer vertical search, NVIDIA DGX Spark brings local LLM inference to desktops, Amazon AgentCore on AWS Bedrock deploys monitored AI agents, Microsoft MarkItDown converts documents to Markdown for LLM pipelines, Nanonets OCR2 adds visual reasoning and multilingual support, and Flint launched an autonomous website builder with $5M funding. LLM Updates: Qwen3-VL expanded from 4B to 235B parameters, video models Veo 3 and Sora 2 traded wins on realism, ServiceNow released a 15B multimodal model, KAIST launched KORMo-10B for Korean-English, Google Gemini 3.0 was leaked, and ChatGPT added Spotify, Canva, and Slack integrations. Research: DiT360 improved panoramic image generation, Phalanx Attention accelerated long-context processing, representation autoencoders aim to replace VAEs, targeted model retraining reduces costs, and IIT Delhi found LLMs struggle with scientific reasoning. Industry: Google's Vizag AI City creates India's largest hub, OpenAI-Broadcom partnership reshapes AI hardware, Walmart integrates ChatGPT shopping, Japan questions OpenAI over Sora 2 anime outputs, Sweden provides free AI access, and 71% of UK workers use unapproved AI. Guides: Embedding model selection for RAG, Qwen3-VL cookbook, thinking tokens explainer, agent security walkthrough, and AI video comparison. Showcases: Sora 2 enables instant content remixes, baby dino AI shifted sentiment, and community contests drew thousands. Discussions: OpenAI Guardrails bypassed by simple injections, reward models miss 25%+ preferences, synthetic data risks model collapse, agents need tool-focused fine-tuning, and resource allocation improves reasoning. Support the show