Send us a text🌍 INAI • The Open AI HubThe 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-wikiInfrastructure & Chips: IBM integrated Groq's LPU inference into watsonx, achieving major speed and cost improvements. Multi-billion chip deals between OpenAI and Nvidia/AMD/Broadcom intensified the compute arms race. Modular expanded GPU support to seven architectures with records on AMD's MI355, while China advanced alternative lithography (SSMB, nanoimprint, multi-beam e-beam). A major AWS outage disrupted OpenAI, Snapchat, Canva, Signal, and Duolingo.Models: Alibaba's Qwen3 launched a trillion-parameter MoE LLM with million-token context. DeepSeek V3.1 delivered strong live trading benchmarks. Claude Sonnet 4.5 and GLM 4.6 climbed web-dev leaderboards. Gemini 3 Pro rumors suggest stronger reasoning; Kimi K2 shows speed/accuracy gains. OpenAI added ChatGPT "selective forgetting" for privacy control. Safety advances included misalignment classifiers, ByteDance's ReSA dataset, and CaRT (teaching models when to act).Tools: Krea 14B open-sourced real-time video (Apache 2.0) streaming at double-digit FPS. DeepSeek OCR handles 100+ languages with context compression. dstack launched GPU dev environments in VS Code/Cursor. TabbyAPI added tensor parallelism. Keras 3 integrated GPTQ quantization. LangChain adopted Model Context Protocol for human-in-the-loop checkpoints.Video & Robotics: Google Veo 3.1 topped leaderboards with frame transitions and object removal. Sora 2 refined moderation. Unitree H2 humanoid debuted with redesigned hip. A Glif-based mobile agent enabled Hollywood-style effects on-the-go.Research: Hugging Face hosts 308GB CommonForms VLM dataset. NVIDIA previewed QeRL for faster RL. Google Gemini identified supernovae with interpretable outputs. DeePFAS detects forever chemicals.Security: Amazon Bedrock Guardrails add protections; Microsoft released cybersecurity benchmark; OpenAI tightened Sora 2 consent protocols.Education: Latent Space released Open Model Pretraining Masterclass. Stanford published end-to-end LM blueprint. Hugging Face launched robotics course.Debates: Developers debate AI code velocity vs review bottlenecks. "AI Operating System" concept emerges. Karpathy emphasized RL for AGI. OpenAI retracted Erdős claims.Support the show

Ver todos los episodios