JAX ADDS NUMPY 2.6 TOP_K, INTEL GPU SUPPORT LANDS IN CI
JAX shipped native top_k alignment with NumPy 2.6.0 while simultaneously landing full OneAPI CI infrastructure for Intel GPU testing across the entire stack.
read --wire →
$ tail -f topics/ai.log
AI / ML news, every morning.
Daily updates from the AI and machine-learning ecosystem - PyTorch, Hugging Face, the Anthropic and OpenAI SDKs, local LLMs, and the agent frameworks.
256 updates across 10 projects this week.
Your week in AI / ML, every Monday.
One calm review of what shipped across AI / ML - the commits, releases, and security advisories that matter. Every Monday, with security advisories same-day. Free, unsubscribe in one click.
We'll start you on the top AI / ML projects - refine anytime. · Read a sample issue →
Recent wires
Meta
REACT NATIVE PLUGS TWO CRITICAL CRASHES ACROSS ANDROID AND iOS
A percentage borderRadius on Android Images and a Debug/Release build swap on iOS are both fixed overnight, clearing two silent production killers from the pipeline.
read --wire →
OpenAI
CODEX HARDENS SECURITY, SDKS MODERNIZE TOOLCHAINS
Codex now prompts users before trusting local directories, closing a prompt injection exposure, while both Node and Python SDKs shift to Oxlint and Oxfmt tooling.
read --wire →
Anthropic
CLAUDE CODE ACTION LOCKS DOWN WORKFLOW PERMISSIONS AND IMAGE PAIRING
The code action now enforces collaborator checks on workflow_run events and pairs downloaded images correctly by asset identifier instead of array position.
read --wire →
AI Agents
LANGCHAIN PLUGS RESOURCE LEAK, CLINE ADDS CONTEXT-WINDOW RECOVERY
LangChain fixed a critical event-loop leak in streaming tracers that was breaking Python 3.14+ tests, while Cline shipped context-window overflow detection and recovery across its SDK.
read --wire →
Local LLMs
OLLAMA REWIRES OPENAI COMPATIBILITY, LLAMA.CPP HARDENS SAMPLING
Ollama reworked its entire streaming protocol to match OpenAI's wire format chunk-for-chunk, unblocking drop-in replacement deployments.
read --wire →
Hugging Face
LEROBOT BREAKS IMPORTS, FIXES TRAINING PIPELINE SILENTLY BREAKING
LeRobot 0.6.1 renames the types module and patches a critical bug where multi-stage training produced all-zero targets when episode annotations were malformed.
read --wire →
JAX HARDENS GPU KERNELS WHILE CLOUD PYTHON PATCHES SUPPLY CHAIN RISK
JAX shipped critical Mosaic GPU optimizations that squeeze performance out of transposed memory transfers, while the cloud SDK locked down six open security vulnerabilities in sqlalchemy-spanner.
read --wire →
Meta
REACT DEVTOOLS KILLS TIMELINE PROFILER, PIVOTS TO PERFORMANCE PANEL
React DevTools is retiring its built-in Timeline profiler and steering users to the browser's native Performance tracks instead, ripping out 74 files of dead code in the process.
read --wire →
OpenAI
GPT-5.5 LANDS ACROSS NODE AND PYTHON SDKS, CODEX HARDENS MCP CONFORMANCE
OpenAI shipped gpt-5.5 model support and tool metadata across both JavaScript and Python SDKs while Codex locked down a full MCP client conformance regression suite to prevent protocol slippage.
read --wire →
Anthropic
CLAUDE CODE AGENT SDK TIGHTENS SKILL VALIDATION, FIXES EVENT FILTERING
The Python agent SDK now rejects malformed skill names that could silently disable tools, while the code action fixes inconsistent timestamp handling across GitHub event types.
read --wire →
AI Agents
LANGCHAIN LOCKS DOWN TOOL ARGUMENT HANDLING, CLINE HARDENS RELEASE PIPELINE
LangChain shipped a burst of tool-handling fixes overnight that close injection gaps in subclasses, while Cline moved desktop signing secrets behind gated environments and killed notification spam.
read --wire →
Local LLMs
DEEPSEEK V4 INFERENCE GETS 50% SPEEDUP; LLAMA.CPP AND VLLM BOTH SHIPPING OPTIMIZED PATHS
DeepSeek V4 speculative decode just went from nice-to-have to standard: llama.cpp landed MTP and DSpark support with roughly 50% speedup [ref:3], vLLM hardened its KV cache coordinator for the same models [ref:7], and Metal got native hyper-connection ops [ref:5].
read --wire →
Hugging Face
LEROBOT FIXES D405 REALSENSE STARTUP HANGS WITH AUTO-RECOVERY
LeRobot's camera stack now survives RealSense D405 hardware lockups that used to require a physical replug.
read --wire →
PyTorch
EXECUTORCH BACKENDS PATCH METADATA BUG, PYTORCH FIXES TEST BLOAT
Qualcomm and Arm backends shipped critical fixes to restore partition metadata and tame test runtime bloat across GPU CI.
read --wire →
JAX SYNCS XLA DEPENDENCY TO LATEST REVISION
JAX pinned its XLA dependency to a fresh upstream revision overnight, pulling in compiler optimizations and fixes across the ML stack.
read --wire →
Meta
REACT NATIVE QUIETLY UPGRADES CODE FORMATTER INTERNALS
One commit landed on React Native master overnight, and it's all infrastructure.
read --wire →
OpenAI
CODEX HARDENS LOGIN FLOW WITH ONBOARDING HINTS AND METADATA PROTECTION
Three critical fixes shipped across auth, data persistence, and usage tracking to prevent thread metadata loss and secure OAuth callback handling.
read --wire →
Anthropic
PLUGIN ECOSYSTEM ROLLS OUT VALIDATED UPDATES ACROSS THREE REPOS
Thirteen automated dependency bumps swept through the plugin pipeline overnight, each validated before merge across official, community, and knowledge-work repositories.
read --wire →
AI Agents
CLINE CHECKPOINTS NOW REWIND RELIABLY ACROSS CLI AND VSCODE
Cline shipped a four-commit fix that makes workspace snapshots actually work as true rewinding across both the VS Code extension and CLI after days of checkpoint failures broke both surfaces.
read --wire →