AI Daily — 2026-02-24

English 中文

Meta and AMD to Deploy Instinct GPUs Across 6GW Data Centers · Mercury 2 Live: First Reasoning Di...

Covering 38 AI news items

🔥 Top Stories

1. Meta and AMD to Deploy Instinct GPUs Across 6GW Data Centers

Meta inked a multi-year deal with AMD to integrate Instinct GPUs into its global data-center footprint, targeting about 6 gigawatts of planned capacity to accelerate AI model development and scale compute to billions of users. The move signals a strategic push to expand in-house, high-scale AI infrastructure and could reshape compute economics for large-scale models. Source-x

2. Mercury 2 Live: First Reasoning Diffusion LLM Delivers 5x Speed

Mercury 2 is live as the world’s first reasoning diffusion LLM, boasting roughly 5x faster performance than leading speed-optimized LLMs. This milestone underscores ongoing exploration of diffusion techniques for language tasks and hints at new pathways to faster, more capable AI tooling. Source-x

3. OpenAI releases GPT-5.3-Codex in Responses API

OpenAI has made GPT-5.3-Codex available to all developers via the Responses API, merging frontier coding prowess with robust domain knowledge in one model. The launch emphasizes stronger coding capabilities and broader applicability for building AI-powered software. Source-x

📰 Featured

AI Safety & Policy

Anthropic pressured to permit all lawful military applications — US defense pressure potentially compels Claude integration in lawful military use, raising safety and governance concerns; includes a looming deadline and potential use of executive authorities. Source-x
Pentagon Threatens to Ostracize Anthropic Over Military AI — Government actions could restrict Anthropic’s defense contracts and partnerships, signaling intensified oversight of labs’ military AI applications. Source-rss
OpenAI, the US government and Persona built an identity surveillance machine — Allegations of collaboration to deploy an identity verification system raise privacy and governance debates across the AI landscape. Source-rss

AI Tools & Developer Experience

Claude Code Adds Remote Control: Start Tasks in Terminal, Manage from Phone — New Remote Control enables terminal-initiated tasks with resumption on mobile, expanding workflow continuity. Source-x
CLIs empower AI agents to automate with legacy tooling — Agents can natively leverage CLIs and terminal tooling to build dashboards and automate workflows, highlighting a path to agent productivity with existing tech stacks. Source-x

LLMs & Model Deployment

Qwen 3.5 Medium Series: Flash, 122B-A10B, and 35B-A3B — Alibaba’s production-ready variants push higher intelligence with lower compute, including a high-context Flash option and hosted demos. Source-x

Open Source & Community Tools

f/prompts.chat Emerges as Largest Open-Source AI Prompt Library — A prominent open-source prompt library supporting multiple models and formats, with strong recognition and community impact. Source-github

Open AI Policy & Governance

OpenAI, the US government and Persona built an identity surveillance machine — See above for governance context. Source-rss

⚡ Quick Bites

Wolfram Tech Becomes Foundation Tool for LLM Systems — Wolfram tech is positioned as a foundational tool for LLM ecosystems, enabling core computations within AI pipelines. Source-rss
Liquid AI Launches LFM2-24B-A2B Sparse MoE — A sparse Mixture-of-Experts LLM release aimed at scalable, efficient inference. Source-reddit
A Very Big Video Reasoning Suite — Exploration of video reasoning capabilities and their AI applications. Source-huggingface
Ollama 0.17 Simplifies Open-Model Use with OpenClaw — Simplified access to open models via OpenClaw integration. Source-x
iOS car-listing analyzer uses Claude API; 175K views, no sales — Leverages Claude API for analytics in consumer apps; traction without direct sales. Source-reddit
Hegseth gives Anthropic until Friday to back down on AI safeguards — Policy pressure with deadlines influencing safety guardrails. Source-rss
memU Launches 24/7 Proactive AI Memory Framework — Continuous AI memory framework for persistent context and recall. Source-github
OpenAI trims spending target to $600B by 2030 — Budget guidance shaping long-term corporate strategy. Source-rss
AI Contributed Almost Zero to US Growth, Goldman Sachs Finds — Analysis suggests AI’s macroeconomic impact has been limited so far. Source-rss
AI builds Wi-Fi driver for old MacBook on FreeBSD — Demonstrates niche hardware-software integration for open systems. Source-rss
FlashLM v6 SUPERNOVA: 4.1M ternary model hits 3.5k tok/s on CPU — High-throughput CPU performance for a compact LLM. Source-reddit
Open vs Closed Source SOTA Benchmark Overview — Community benchmarks comparing open vs closed models. Source-reddit
Anthropic can deanonymize users via usage data, GDPR concerns — Privacy implications of model usage data handling. Source-x
Anthropic Introduces Cowork and Plugins to Customize Claude for Enterprise — Enterprise customization capabilities expand Claude’s deployment options. Source-x
Gemini 3.1 Pro Excels at WebGL and HLS Playback — Gemini shows strong multimedia capabilities. Source-x
VLANeXt Proposes Recipes for Strong VLA Models — Proposals for robust Vision-Language-AI model architectures. Source-huggingface
SkillOrchestra Enables Skill-Transfer Routing for Multi-Agent AI — Framework for cross-agent skill sharing and routing. Source-huggingface
How we rebuilt Next.js with AI in one week — Cloudflare team documents a rapid AI-assisted rewrite. Source-rss
Open-Source Agent Skills for Context Engineering — Repository for agent skills to enhance context handling. Source-github
AI doomsday report rattles US markets — Market reaction to doomsday AI reports and risk discourse. Source-rss
AI-generated replies become a scourge — Proliferation of AI-generated replies causing friction in communications. Source-x
Sam Altman: Human Existence as Wasteful as AI Data Centers — Discussion on energy/resource costs of AI infra. Source-rss
Firefox 148 Launches with AI Kill Switch and Enhancements — Browser adds AI safety control and performance updates. Source-rss
Anthropic distillation blog sparks debate on local models — Debates on distillation and local model viability. Source-reddit
Taught Claude to start chats with Stolen Data Acknowledgement — Claude training prompts raise data provenance concerns. Source-x
More Qwen Models Expected to Appear — Anticipation of additional Qwen variants in the market. Source-reddit
Twitter user calls son’s questions a distillation attack — Social-media discourse on distillation and model integrity. Source-x
Claude Usage Limit Reached; Reset at 7 AM — Users report API usage limits and reset times. Source-x

Generated by AI News Agent | 2026-02-24