today's signal · no scroll
the brief
Tooling shipped across the stack: Claude Code turns Auto mode on by default, Vercel adds Tree/Waterfall traces, Cloudflare sharpens tiered caching, and Next.js canary advances. The headline is Apple suing OpenAI, as OpenAI’s safety leadership also shifts. Research pushed on decoding and quantization, while “GhostLock” flags a long-lived Linux kernel risk.
Researchers detail a long‑lived stack use‑after‑free affecting Linux distributions; monitor distro advisories, update kernels promptly, and review hardening to reduce exposure.
GhostLock, a stack-UAF that has existed in ALL Linux distributions for 15 years — Article URL: https://nebusec.ai/research/ionstack-part-2/ Comments URL: https://news.ycombinator.com/item?id=48864969 Points: 42 # Comments: 9
Anthropic’s Claude Code v2.1.207 now defaults Auto mode on Bedrock, Vertex AI, and Foundry, with stability and consent fixes aimed at smoother agentic coding workflows.
v2.1.207 — What's changed Auto mode is now available without CLAUDE_CODE_ENABLE_AUTO_MODE opt-in on Bedrock, Vertex AI, and Foundry; disable via disableAutoMode in settings Fixed the terminal freezing and keystrokes lagging while streaming responses containing very long lists, tables, paragraphs, or code blocks Fixed remote managed settings from a non-interactive run (claude -p, the SDK) being permanently recorded as consented without ever showing the security consent dialog Fixed spurious pr...
New Tree and Waterfall views in Vercel Logs visualize span hierarchy, critical path, and timing inline, tightening the loop on performance debugging without leaving logs.
Traces now support Tree and Waterfall views — now support Tree and Waterfall views. See span hierarchy, critical path, and timing without leaving the log entry.Traces in Vercel Logs The Tree view groups spans by parent and child. On load, spans are sorted by duration so the slowest appear first. The Waterfall view places every span on one chronological axis, showing what ran in parallel and where time was lost. Read the to learn more.Tracing documentation Read more
Smart Tiered Cache now supports customer-provided cloud region hints, improving upper-tier selection for AWS, GCP, Azure, and Oracle origins to trim latency and egress.
Improving Smart Tiered Cache for Public Cloud Regions — Smart Tiered Cache allows for precise upper tier selection for origins hosted on AWS, GCP, Azure, and Oracle Cloud with customer-provided cloud region hints.
The latest 16.3.0-canary updates add LightningCSS custom media support, clarify AI-assisted contribution policy, and refine staged rendering internals and docs.
v16.3.0-canary.83 — Misc Changes Turbopack tests: remove assertions that duplicate webpack results: #95688 Fix support for custom-media-queries in LightningCSS: #95689 Clarify AI-assisted contribution policy in PR template and AGENTS.md: #95629 docs: add Building guide: #94999 docs: add incremental adoption path to Cache Components migration guide: #95325 [PPF] Sync IO is only allowed in the dynamic stage: #95384 refactor: remove unnecessary switches from StagedRenderingController: #95383 Kee...
Apple alleges former employees took confidential information to OpenAI, escalating platform–LLM vendor tensions and raising stakes for partnerships and hiring practices.
Apple sues OpenAI, accuses ex-employees of stealing trade secrets — https://www.macrumors.com/2026/07/10/apple-sues-openai/ Comments URL: https://news.ycombinator.com/item?id=48865019 Points: 618 # Comments: 297
OpenAI’s head of safety departs as research and safety teams integrate; Mia Glaese steps in as VP, signaling ongoing reorg of governance and risk functions.
OpenAI's head of safety, Johannes Heidecke, is leaving as OpenAI integrates its research and safety teams; Mia Glaese will become VP of research and safety (Maxwell Zeff/Wired) — Maxwell Zeff / Wired: OpenAI's head of safety, Johannes Heidecke, is leaving as OpenAI integrates its research and safety teams; Mia Glaese will become VP of research and safety — Johannes Heidecke's departure comes as OpenAI tries to further integrate its research and safety teams.
Paper introduces confidence-scheduled speculative decoding with semi‑autoregressive generation to reduce latency while preserving quality, offering another path to faster LLM serving.
[10/30] 414 Likes, 2 Comments, 2 Posts 2607.05147, cs․AI | cs․CL, 06 Jul 2026 🆕DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation Xin Cheng, Xingkai Yu, Chenze Shao, Jiashi Li, Yunfan Xiong, Yi Qian, Jiaqi Zhu, Shirong Ma, Xiaokang Zhang, Jiasheng Ye, Qinyu...
TMLR work shows learning from important tokens improves quantized LLMs, pointing to token-importance–aware training as a practical route to regain lost precision.
RSQ: Learning from Important Tokens Leads to Better Quantized LLMs Yi-Lin Sung, Prateek Yadav, Jialu Li, Jaehong Yoon, Mohit Bansal Action editor: Gustavo Carneiro https://openreview.net/forum?id=kBezrKXHVS #attention #quantization #quantize
A systematic study contrasts spatial versus frequency masking for continual test-time adaptation, mapping stability–plasticity trade-offs that guide robust deployment choices.
Family Matters: A Systematic Study of Spatial vs. Frequency Masking for Continual Test-Time Adapt... Chandler Timm Cagmat Doloriel, Yunbei Zhang, Yeonguk Yu et al. Action editor: Stephen Lin https://openreview.net/forum?id=pBI64qNXHp #adaptation #masking #masked
Simon Willison spotlights Nilay Patel’s argument that realistic AR needs near‑eye cameras and offboard compute—underscoring unavoidable privacy, power, and UX trade‑offs.
Quoting Nilay Patel — <blockquote cite="https://youtu.be/v4vkwUf4AMw?t=2427"><p>The reality is to make augmented reality glasses, you need to put a camera next to your eyes that is continuously recording everything you see and processing that to put information over it.</p> <p>There is not another way around it. And there's certainly not a chip that can fit in the stem of a glasses that is both powerful enough and power miserly enough to do that in real time.</p> <p>You have to send that data...