TypeSafe AI introduces Jev, a new class of System One Models designed for machine automation rather than chat. Unlike RLHF-trained language models, Jev produces typed decisions with calibrated confidence estimates, enabling reliable autonomous workflows at significantly lower cost and latency.
Lumai has unveiled an optical computing system for AI inference that uses 3D optical components to achieve 10-100x performance and efficiency gains over traditional silicon accelerators, with 90% power reduction and 10% of typical total cost of ownership while supporting large language models at datacenter scale.
Data broker Acxiom compiles detailed dossiers on billions of people worldwide, including inferences about their weight, alcohol consumption, medical procedures, and spending habits. Consumer Reports obtained over 100 personal files from Acxiom to test state privacy laws, revealing the company makes predictions across 3,000+ categories and sells data to major companies like Disney, GM, and financial institutions.
A developer releases Qwen-2.5-1B-RLCD, an open-source model enabling 5x faster on-device JSON inference without new training. The same developer also announces Jev, a frontier AI model developed over 2 years using a new training method (RLCD) that claims 20-200x performance improvements.
Apple is considering developing an AI server using M8 chips and Nvidia networking hardware for external customers, potentially returning to the server market it exited in 2011. The system would target companies running AI inference workloads, with a potential 2029 release date, though the project remains unfinalized and could be canceled. The initiative reflects growing AI hardware demand and signals a closer working relationship between Apple and Nvidia.
A developer created a minimal operating system powered by Alibaba's Qwen 3.8 27B model running at 1,950 tokens per second on Cerebras hardware, where applications are generated on-demand through model inference rather than stored on disk. The system demonstrates rapid app compilation, with a calculator loading in 11 seconds and a paint studio in 10 seconds, at a cost of $0.99 per million input tokens.