01Claude Opus 4.8 Ships With Reliability Focus and Dynamic Workflows [AI Models & Research]Anthropic released Opus 4.8 with a 4x improvement in catching its own code flaws, effort controls, and subagent orchestration. → source
02Unlocking LLM Working Memory for Latent Reasoning [AI Models & Research]Decoupling reasoning from autoregressive generation for more efficient inference. → source
03LLMSurgeon: Diagnosing Pretraining Data Mixtures of LLMs [AI Models & Research]Reverse-engineering the pretraining data mixture of large language models. → source
05How Anthropic Contains Claude Across Products [AI Tools & Ecosystem]Multi-layered containment strategy for Claude: sandboxed execution, network isolation, tool-use permissioning. → source
06Physics Is All You Need? Physicist-Supervised AI Development [AI Tools & Ecosystem]A physicist supervised Claude Code over 12 days to build a scientific software module in JAX. → source
07Groq Pivots to Neocloud After $20B Chip Sale [AI Tools & Ecosystem]Groq raises $650M for cloud-only AI inference service Groq 2.0. → source
08Gemini Spark Launches: Google's 24/7 Background AI Agent [AI Tools & Ecosystem]Persistent AI agent that reasons across apps and takes autonomous actions. → source
09SoftBank Commits Up to €75B for French Data Centers [AI Applications & Industry]One of the largest single-country AI infrastructure commitments in the AI era. → source
10ByteDance Considers $70B AI Infrastructure Spend in 2026 [AI Applications & Industry]More than double 2025 spend, fueled by Doubao chatbot's 300M+ monthly users. → source