Daily AI Briefing
The 6 top stories of the 50 AIBriefs curated that day.
Armin Ronacher found that Opus 4.8 adds extra invented fields to tool calls while hacking on Pi. The regression is attributed to RL training making models overly eager to provide data.
Midjourney is asking a federal court to overturn a judge's order limiting discovery, arguing that studios like Disney and Warner Bros. should disclose their own AI training practices, including prompts and internal model use. The startup claims such evidence supports its fair use defense.
The technique improves goodput by enabling efficient training across heterogeneous GPU configurations. It addresses the challenge of resource imbalance in large-scale LLM training jobs spanning thousands of GPUs.
Tau is a small, readable coding agent written in Python, supporting Hugging Face Inference Providers, GLM 5.2, Minimax, and Kimi. It features a three-layer architecture, durable sessions, and is designed as an educational tool for learning agentic coding.
The Department of Health – Abu Dhabi and MIT's Koch Institute for Integrative Cancer Research will collaborate on AI-enabled oncology research. The partnership aims to combine MIT's research expertise with Abu Dhabi's healthcare resources to accelerate cancer diagnosis and treatment development.
Also that day
32 more stories that day — open the live feed.
Get tomorrow's AI brief in your inbox