9th September - AI News Daily - Cognition Raises $400M at $10.2B Valuation as Alibaba Launches 1T-Parameter Model

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 — 09 Sept 2025: Comprehensive Summary 

Top Highlights:

Cognition secures $400M at $10.2B valuation to expand autonomous coding agentsHugging Face releases FinePDFs (3T-token dataset) and revives SOTA leaderboardsGoogle Gemini faces backlash over conversation analysis defaultsAlibaba debuts Qwen-3-Max-Preview with 1T parametersNVIDIA projects $3-4T AI infrastructure market despite China uncertainties

New Tools: NVIDIA ModelOpt (open-source optimization toolkit), LangGraph + LangChain v1 Alpha (streamlined multi-agent workflows), LlamaIndex updates, LavaMoat + FastMCP 2.12 (security improvements), Glass for iOS (clinical decision support), and Vercel's coding platform

LLM Updates: Kimi K2-0905 exceeds 90% on Roo Code, breakthrough inference speeds from Groq and Meta, Alibaba's 1T-parameter model, NVIDIA's findings on compact models, Qwen3-ASR's multilingual transcription, and OpenAI acknowledging GPT-5 hallucinations

Research: Hugging Face's PDF dataset, parallel reasoning gains, Penn Wharton's AI productivity forecast, OpenAI's hallucination analysis, and DeepMind's embedding retrieval insights

Industry & Policy: Cognition's funding, NVIDIA's market projections, Google Gemini privacy issues, Warner Bros vs. Midjourney lawsuit, policy developments (Anthropic, Perplexity), and automotive AI integrations

Tutorials: Fine-tuning techniques, preference optimization approaches, agentic RL survey, PyTorch AOT compilation, and MLX + Qwen2 serving

Demos: Google/Intrinsic's robot coordination, DeepMind Recomposer for audio editing, KradleAI's Minecraft model testing, MatAnyone's video matting, open-source robotics advances, and visual PDF search innovations

Key Discussions: Evaluation metrics standardization, training philosophy debates, hallucination discourse, retrieval system approaches, systems experimentation cost wins, and long-term AI forecasts including 1000× larger training runs by 2030 and widespread work automation by 2035
 

Support the show