The through-line
Infrastructure for AI is maturing across multiple layers: execution environments (Python Workers), model efficiency (pruning, sparse attention, routing), and operational tooling (tokenization, agent memory). Meanwhile, consumer AI disappointments (Apple Siri) and tooling gaps (identifying LLM call sites) expose the gap between capability and delivery.
Issue signature
The Wire
- Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem
- Python Workers are now generally available
- 5 Companies Using NVIDIA AI for Clean Energy
- RBS-Attention: Radius-Bounded Sparse Prefill for Long-Context Large Language Models
- Attention-Aware Routing: Coupling Routing and Attention in MoEs
- How V7 gives AI agents institutional memory
- tokenizers v1: encode, decode and scaling, measured
- Can CEO John Ternus find Apple’s next big thing?
- iPhone owners can now submit a claim in Apple’s $250 million Siri AI lawsuit
- You cannot tell which parts of your software should stop calling an LLM. My Jev guide has a prompt that scans your projects and names them.