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

📰 AI NEWS DAILY SUMMARY — 11 SEPT 2025

TOP HIGHLIGHTS: Microsoft integrates Claude into Office 365 alongside OpenAI models; Oracle reportedly secures $30B OpenAI cloud deal amid tightening inference capacity; Enterprise AI agents launched by Adobe, DeepL, and ServiceNow; U.S. Census reveals first decline in business AI adoption since 2023; Gemini 2.5 Pro leads new factuality benchmarks while open models gain in multilingual and robotics tasks.

NEW TOOLS: Enterprise agents from DeepL, Adobe, and ServiceNow focus on workflow automation with governance; Together AI enables fine-tuning for 100B+ models with 131k context; Google releases production-ready Genkit Go 1.0; DSPy ecosystem expands with Rust port and stateful conversations; Developer tools advance with Chroma, Qodo Aware, and Swarm; Hugging Face launches LeRobot and OpenVLA 7B for physical AI prototyping.

MODEL PROGRESS: Benchmarks show Gemini 2.5 Pro leading in factuality and audio tests; BackendBench demonstrates models passing half of PyTorch operator tasks; Notable releases include mmBERT, ModernBERT, EmbeddingGemma, K2-Think 32B, and DeltaNet; OpenVLA 7B exceeds larger models in robotics manipulation; Infrastructure advances with deterministic inference and multimodal embeddings; Evaluation methods reveal benchmark vulnerabilities.

RESEARCH BREAKTHROUGHS: MIT advances chemical reaction modeling; DeepMind and Imperial study antibiotic resistance via "pirate phages"; Harvard's PDGrapher targets gene-drug combinations for neurodegenerative diseases; Cell-therapy AI reverses disease states; AI reduces radiology diagnostic errors; Green AI research weighs sustainability impacts.

INDUSTRY MOVES: Microsoft diversifies with Claude in Office 365; Oracle-OpenAI deal worth ~$30B; White House hosts tech leaders on AI policy; OpenAI forms Applied Evals team; Anthropic expands Fellows Program; Major funding rounds for Replit ($250M), Cognition ($400M), Mistral ($14B valuation), and Standard Fleet ($13M); Chile advances ethical AI legislation; OpenAI faces legal challenges; Business AI adoption declines for first time since 2023.

PRACTICAL KNOWLEDGE: New resources on KV cache compression, robust RAG implementation, evaluation design, foundational NLP training, deterministic LLM inference, and cloud optimization strategies.

DEMOS & DISCUSSIONS: Terminal-based agent climbs SWE-bench leaderboard; Claude generates complete financial models; Seedream v4 enables "infinite painting"; VLMs run on edge devices; MCP setups control robots and generate images; Research demos phone fingerprinting and affordable robot dexterity; Debates emerge on AI-authored political speeches, humans-only social networks, requirements for "true understanding", agent engineering patterns, and market predictions through 2028.

Support the show