🖥️ IBM z17 Mainframe 🖥️
7.5x more AI performance

IBM has unveiled the z17, its most powerful mainframe computer designed specifically for advanced AI workloads in enterprise environments. The system builds on IBM's traditional strengths of security and reliability for mission-critical operations while adding enhanced AI capabilities to support large language models, assistants, and agents. The z17 delivers 7.5x more AI acceleration using 5.5x less power than traditional systems, making it ideal for enterprises that need to implement AI in their core operations without compromising on security or reliability.

🧠 Deep Cogito Open AI Models 🧠
A path to general superintelligence

Deep Cogito has released a new family of open source LLMs ranging from 3B to 70B parameters, outperforming the best available open models of the same size, including those from LLaMA, DeepSeek, and Qwen. Their 70B model even outperforms the newly released Llama 4 109B MoE model. Trained using Iterated Distillation and Amplification (IDA), these models can both answer directly or self-reflect before responding. Deep Cogito plans to release larger models up to 671B parameters in the coming months, with all models remaining open source.

🤖 Stanford AI Index 2025 🤖
AI becomes smarter and cheaper

Stanford University has released its comprehensive 2025 AI Index Report, analyzing AI's rapid evolution across the industry. The 400+ page report reveals AI became substantially smarter and cheaper to use in 2024, with performance on benchmarks skyrocketing (e.g., SWE-bench solving increased from 4.4% to 71.7% in one year). Usage costs plummeted to 1/280th of 2022 prices, while business adoption soared to 78% (up from 55%). However, training costs are escalating dramatically, with future models potentially costing $1B+ to develop.

INTERESTING NEWS

🤝 AI agents raise money for charity 🤝

Sage Future, a 501(c)(3) nonprofit backed by Open Philanthropy, launched an experiment tasking four AI models in a virtual environment with raising money for charity. The models — OpenAI's GPT-4o and o1 and two of Anthropic's newer Claude models — had freedom to choose which charity to fundraise for and how to drum up interest. In around a week, they raised $257 for Helen Keller International. The agents demonstrated impressive autonomous capabilities, coordinating with each other, creating documents, researching charities, and even setting up social media accounts.

🚗 Waymo's data collection plans 🚗

Waymo has drafted a new privacy policy that indicates plans to use in-cabin camera footage for AI model training and personalized advertising. This policy reveals how the autonomous vehicle company intends to utilize passenger data captured within their vehicles. The approach has sparked fresh concerns among privacy advocates regarding data collection practices in self-driving technology. Currently under review, this draft policy highlights the evolving relationship between AI development and personal privacy in transportation.