AI Daily Brief — 21 May 2025
OpenAI acquires Jony Ive's io in a $6.5B all-equity deal — OpenAI's largest ever. 9-minute Altman-Ive teaser film hints at a screenless 'family of devices.' Mistral…
OpenAI acquires Jony Ive's io in a $6.5B all-equity deal — OpenAI's largest ever. 9-minute Altman-Ive teaser film hints at a screenless 'family of devices.' Mistral…
Google I/O 2025: Gemini 2.5 Pro tops LMArena; Gemini app 400M MAU; Deep Think mode; Veo 3 ships with synchronized audio; Imagen 4 at 2K; AI…
Microsoft Build 2025 opens with 50+ announcements — Copilot Tuning, autonomous GitHub Copilot Coding Agent, Microsoft Discovery (200-hour datacenter coolant), Windows AI Foundry, NLWeb, Entra Agent…
Eve of the densest enterprise-AI week of Q2. Jensen Huang takes the Computex stage Sunday evening PT (Monday morning Taiwan) — NVLink Fusion, DGX Spark and…
Quiet Saturday. xAI's Grok system-prompt incident from earlier in the week keeps fueling the news cycle — France 24 and CNBC both run analysis pieces. CNBC…
Trump concludes the Gulf trip — $200B in new US-UAE deals plus an accelerated $1.4T UAE investment in US AI/tech. Total trip deal value approaches $2T.…
Trump and MBZ unveil a 5GW UAE-US AI Campus in Abu Dhabi — largest AI infrastructure project outside the US. G42 builds, OpenAI + Oracle operate,…
OpenAI rolls GPT-4.1 and GPT-4.1 mini into ChatGPT — GPT-4.1 mini replaces GPT-4o mini as the Free-tier default. Trump lands in Doha with the same CEO…
Megaday for the Gulf AI pivot. Trump lands in Riyadh, signs $600B US-Saudi pact. BIS formally rescinds the AI Diffusion Rule. NVIDIA-Humain: 18,000 GB300 Blackwells for…
Commerce signals Biden AI Diffusion Rule rescission. US-China 90-day tariff truce — US tariffs on Chinese imports drop 145% → 30%. OpenAI ships HealthBench (262 physicians,…
Pope Leo XIV delivers his first Sunday Regina Caeli at St. Peter's Square — ~100,000 attendees. Three days after election; the AI-as-papacy-priority thread from yesterday continues…
Pope Leo XIV's first formal address to the College of Cardinals explicitly frames AI as 'another industrial revolution' posing 'new challenges for the defense of human…
Reuters reports NVIDIA will ship a downgraded H20 to Chinese cloud customers (Alibaba, ByteDance, Tencent) as early as July to comply with US export thresholds. Trump…
Cardinal Robert Prevost elected Pope Leo XIV — first US-born pope, name signaling AI as a defining issue for his pontificate. Senate AI hearing: Altman, Lisa…
Mistral launches Medium 3 — claimed ≥90% of Claude 3.7 Sonnet at $0.40/$2.00 per M tokens — plus Le Chat Enterprise. OpenAI names Instacart CEO Fidji…
OpenAI agrees to acquire Windsurf for ~$3B — its largest acquisition to date. NVIDIA + ServiceNow unveil Apriel Nemotron 15B open-source reasoning model for enterprise agents.…
OpenAI reverses course on for-profit conversion. The nonprofit will retain control; the for-profit LLC converts to a Public Benefit Corporation. Bret Taylor publishes board chair letter;…
Quiet Sunday. The dominant AI-adjacent story is the White House's AI-generated Trump-as-Jedi Star Wars Day post — fans noting the red lightsaber canonically belongs to Sith…
Quiet Saturday. Industry digests Qwen3, DeepSeek Prover-V2, Llama 4 and the OpenAI sycophancy postmortem. The AI Diffusion Rule is still scheduled to take effect May 15;…
OpenAI publishes the long-form sycophancy postmortem 'Expanding on what we missed' — explicit sycophancy evaluation will be added to launch process; behavioural problems will be launch-blocking;…
Anthropic ships Integrations (10+ partners) and 45-minute Research mode. OpenAI fully rolls back the GPT-4o sycophancy update. ChatGPT shopping search rolls out worldwide — 1B+ web…
DeepSeek releases Prover-V2 in 7B and 671B-MoE variants. 88.9% pass-rate on MiniF2F-test SOTA. Visa unveils Intelligent Commerce with Anthropic, IBM, Microsoft, Mistral, OpenAI, Perplexity, Samsung and…
Meta LlamaCon: Llama API preview, standalone Meta AI app on Llama 4, Llama Guard 4 + Prompt Guard 2 + LlamaFirewall + CyberSecEval 4, Cerebras serves…
Alibaba launches Qwen3 — eight open-source hybrid-reasoning models (six dense 0.6B-32B, two MoE: 30B-A3B and flagship 235B-A22B) under Apache 2.0. First lab to ship togglable 'thinking…