10th October - AI News Daily - OpenAI Inks Nvidia-AMD Chip Deals as AI Infrastructure Era Tops $1T

Send 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 secured major chip partnerships with Nvidia and AMD, signaling AI infrastructure could hit $1T+ annually. Google launched Gemini Enterprise and Amazon debuted Quick Suite, intensifying competition with Microsoft Copilot. Frontier models achieved breakthroughs: GPT-5 Pro leads ARC-AGI, Gemini 2.5 Deep Think tops FrontierMath, Claude 4.5 excels at sustained execution. Industry consolidation continues as Elastic acquires Jina AI and Weaviate partners with Confluent. ChatGPT now serves 800M weekly users.

New Tools: Google Gemini Enterprise ($21/user) offers secure agent-building. Amazon Quick Suite provides AI-first analytics and automation. OpenAI GPT-5 API adds function calling and web search. Weaviate Query Agent introduces agentic RAG. Hugging Face Hub ships custom domains, GGUF edits, and MCP-UI support. Mem0 adds persistent agent memory; FastMCP enables one-click deployment.

LLM Updates: GPT-5 Pro tops ARC-AGI; Gemini 2.5 Deep Think sets FrontierMath record. Claude Sonnet 4.5 runs two-hour uninterrupted tasks. AI21 Jamba Reasoning 3B leads small-model instruction; 7M-parameter Tiny Recursion Model excels. Radical Numerics releases 30B sparse-MoE diffusion model. Microsoft UserLM-8B simulates user behavior. Qwen3-30B hits 473 tokens/sec; OpenAI Codex surpasses Claude Code.

Research: Latent Diffusion and GLASS Flows advance reasoning efficiency. First-token steering and Exploratory Annealed Decoding improve control. MS-SSM scales multi-resolution learning. Attention sinks and compression valleys clarify transformer internals. LoRA-based RL matches full-parameter training; RLAD and bootstrapped methods enhance robustness. Safety work includes inoculation prompting and backdoor detection.

Industry: OpenAI-Nvidia-AMD deals reshape semiconductor supply chains. Elastic-Jina AI and Weaviate-Confluent consolidate vector search. OpenAI urges EU AI competition enforcement. China imposes rare-earth export controls. Security concerns: Sora impostor apps, AI girlfriend data leaks, Gemini injection risks.

Tutorials: DeepMind releases gemma3-270m fine-tuning Colab. Weaviate+DSPy sessions show 20x cost reduction. Sessions cover LLM history, Netflix ML interviews, Stanford alignment lectures, and training sparse models on consumer GPUs.

Showcases: Genie 3 generates playable worlds. Marketing twin agents automate SEO workflows. Smart Cellular Bricks blend robotics with construction. Claude 4.5 builds complete Datasette plugin. Yupp AI demonstrates visual SVG prompting.

Discussions: Calls for reproducibility in robotics. Safety debates on bias measurement, backdoors, data poisoning. Evaluation reliability questioned. New concepts: COLMs, early-token steering, RL critiques. Predictions: LLMs may outperform elite forecasters by 2026.

Support the show

🌍 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