Blog — Page 54
Engineering notes, case studies, and lessons from the field.
Huawei's Ascend 950PR Wins ByteDance and Alibaba Orders as China Ditches Nvidia
ByteDance and Alibaba plan to place orders for Huawei's new Ascend 950PR chip after testing showed improved CUDA compatibility and inference speeds. Huawei targets 750,000 units shipped in 2026.
NVIDIA ProRL Agent Brings Rollout-as-a-Service to Multi-Turn LLM Training
NVIDIA released ProRL Agent on March 27, a decoupled reinforcement learning infrastructure that separates rollout generation from the training loop — designed to scale multi-turn LLM agent training to production.
GitHub Will Train AI on Your Code by Default Starting April 24 — Here's How to Opt Out
GitHub announced on March 25 that interaction data from Copilot Free, Pro, and Pro+ users will be used to train AI models starting April 24, 2026, unless users actively opt out. The move affects millions of developers and has triggered a wave of backlash in developer communities.
Google's AlphaEvolve Recovers 0.7% of Global Compute and Cracks a 55-Year-Old Math Problem
Google DeepMind's AlphaEvolve — a Gemini-powered coding agent that pairs large language models with evolutionary algorithms — is running in production at Google, continuously recovering computing resources and accelerating Gemini training. It also became the first AI system to beat Strassen's 1969 matrix multiplication algorithm.
Model Context Protocol Hits 97 Million Monthly Downloads — AI's New Infrastructure Layer
The Model Context Protocol reached 97 million monthly SDK downloads in March 2026, up from 2 million at launch 16 months ago. Every major AI provider now ships MCP-compatible tooling, and over 5,800 community servers cover databases, CRMs, and cloud platforms — making MCP the de facto standard for AI agent integration.
OpenAI Codex Ships Broad Release with First-Class Plugins and Multi-Agent Overhaul
OpenAI's Codex coding agent received its biggest update yet on March 27, 2026, bringing first-class plugin support, a redesigned multi-agent architecture, and a fully capable app-server TUI. The release signals OpenAI's push to make Codex the default interface for professional software development.
Arm Builds Its First Chip in 35 Years — and Meta Is First in Line
Arm launched the AGI CPU, its first in-house silicon, ending a 35-year run as a pure IP licensor. The 136-core, 3nm chip targets AI inference in data centers, with Meta as the launch customer and OpenAI, Cloudflare, and SAP close behind.
Anthropic Gives Claude Code an Autopilot Mode — With Guardrails
Claude Code's new auto mode lets the AI execute multi-step tasks with fewer interruptions, using built-in safeguards to block risky actions before they run. Anthropic also shipped computer use in Cowork, letting Claude operate your Mac directly.
Cursor Built Composer 2 on a Chinese Open-Source Model — and Didn't Say So
Cursor launched Composer 2 as a proprietary frontier coding model. Within hours, a developer found it was built on Moonshot AI's Kimi K2.5. Cursor admitted the omission — but the licensing implications are still unresolved.
Page 54 of 61 · 549 articles