Prompt Notes #7 - Weekly AI Digest
AI goes fast. Helping you (and myself) catch up.
Welcome to the 7th edition of Prompt Notes, your new favourite weekly AI digest.
What caught my attention
Nvidia had a strange week.
Jensen Huang took the GTC stage, projected $1 trillion in AI chip sales, and spent part of his keynote on OpenClaw before announcing NemoClaw, a reference stack to make it enterprise-ready.
The gentleman who made billions selling shovels is now paving the road to the gold mine. Maybe the “AI commodity threat” is real, after all? :)
Meanwhile, DLSS 5 was announced, and gamers hated it (See my Patch Notes #64). Uncanny visuals, developer pushback. Strange place to be for a company that used to be the unambiguous hero of PC gaming - before crypto made their cards unaffordable, before AI made them a Wall Street story. The original fanbase got priced out somewhere along the way.
Both things happened in the same week. That’s the Nvidia Arc in miniature.
TLDR
Nvidia GTC: Jensen Huang projected $1 trillion in AI chip sales through 2027, announced NemoClaw as an enterprise-ready open-source stack for OpenClaw, and declared every company needs an agentic AI strategy. (TechCrunch AI)
Chinese AI company MiniMax released M2.7, which reportedly participated in its own development through autonomous optimization loops. (The Decoder)
Google revamped Stitch into a full AI design platform with infinite canvas, voice interaction, and MCP integration with Claude Code and Cursor — free tier, Figma exports. Figma shares dropped 8%. (Winbuzzer)
The US Defense Department established Palantir’s Maven AI system as a program of record for integration into combat operations. (Bloomberg Technology).
NOTE: new Terminator script looking mighty fine.
OpenAI began testing ads in ChatGPT. (CNBC Tech)
Rest of the news
Models & Research
Chinese AI company MiniMax released M2.7, which reportedly participated in its own development through autonomous optimization loops. (The Decoder)
Xiaomi launched three MiMo AI models to power AI agents, robots, and voice capabilities that can control software and shop in browsers. (The Decoder)
Qualcomm AI Research developed a modular system compressing reasoning chains by 2.4x for on-device language models. (The Decoder)
Andrej Karpathy reported an autonomous agent optimized his training setup overnight, finding improvements missed despite his two decades of experience. (The Decoder)
OpenAI Chief Scientist Jakub Pachocki now uses AI for experiments that previously took a week, though he does not trust it for designing complex systems. (The Decoder)
Terence Tao stated AI reduces idea generation cost in mathematics to near zero but shifts the bottleneck to verification, requiring new infrastructure. (The Decoder)
OpenAI is refocusing research on building a fully automated AI researcher agent for complex problems. (MIT Tech Review)
Products & Tools
Cursor released Composer 2, its second-generation AI model for software development built on Chinese open-source Kimi K2.5 to match leading models at lower costs. (The Decoder)
OpenAI published a prompting playbook to help front-end designers achieve better results from GPT-5.4 when building websites and apps while preventing fallback to generic designs. (The Decoder)
OpenAI plans to merge ChatGPT, Codex, and Atlas browser into a single desktop superapp after deeming its multi-product strategy flawed. (The Decoder)
Gemini’s new task automation feature on Pixel 10 Pro and Galaxy S26 Ultra enables the AI to use a limited subset of apps like food delivery and rideshare services. (The Verge AI)
Adobe expanded Firefly to bundle over 30 AI models and allow training custom styles on user images. (The Decoder)
WordPress.com introduced AI agents that can write and publish posts. (TechCrunch AI)
Google expanded Universal Commerce Protocol with shopping cart, catalog, and loyalty features for AI agents. (The Decoder)
Google pulled back on browser AI development as coding agents gain preference in industry. (The Decoder)
Google revamped Stitch into a full AI design platform with infinite canvas, voice interaction, and MCP integration with Claude Code and Cursor — free with 350 monthly generations, exports to Figma. Figma shares dropped 8% on the news. (Winbuzzer)
OpenAI began testing ads in ChatGPT. (CNBC Tech)
OpenAI’s first online shopping attempt with Instant Checkout faced inaccuracies in item information and merchant onboarding difficulties with partners like Etsy, Walmart, and Shopify. (CNBC Tech)
Walmart’s test showed ChatGPT checkout converted 3x worse than its own website. (MarTech)
Microsoft reduced Copilot AI entry points on Windows in apps like Photos, Widgets, and Notepad. (TechCrunch AI)
OpenAI acquired Astral, maker of popular Python developer tools, for its Codex AI coding platform. (The Decoder)
iNaturalist app identifies plant, animal, and insect species while providing data to scientists on biodiversity, species decline, and habitat loss. (CBS News Tech)
Infrastructure & Compute
Elon Musk announced Terafab, a chip plant in Austin jointly run by Tesla and SpaceX to produce chips at scale for robotics, AI, and space-based data centers. (The Verge AI)
Nvidia CEO Jensen Huang delivered a GTC keynote projecting $1 trillion in AI chip sales through 2027 and declaring every company needs an OpenClaw strategy. (TechCrunch AI)
Nvidia’s GTC conference demonstrated industry confidence despite investor fears of an AI bubble, with Wall Street remaining unconvinced. (TechCrunch AI)
Nvidia struck a multi-year deal with AWS to deliver around one million AI processors by end of 2027, including GPUs, networking, and inference chips. (Capacity Global)
AWS invited a tour of its Trainium chip lab following a $50 billion investment announcement in OpenAI. (TechCrunch AI)
SoftBank plans a massive AI data center on federally owned land in Ohio powered by $33 billion in natural gas-fired electricity by end of decade. (Bloomberg Technology)
Alibaba shipped 470,000 AI chips and predicts $100 billion in future cloud and AI revenue. (Data Center Dynamics)
Bezos-backed Blue Origin filed for approval to place data centers in space. (Data Center Dynamics)
Data centers are forcing utilities to adopt a new playbook for renewable energy deployment amid mounting pressure. (Data Center Dynamics)
Power shortages for AI data centers create investment opportunities in energy tech. (TechCrunch AI)
Saudi Arabia’s Edarat secured a data center contract from a major regional bank and was selected for AI data center design and supervision services. (Data Center Dynamics)
Supermicro co-founder resigned from board after China GPU smuggling charge involving $2.5 billion in servers, with shares down 33%. (Data Center Dynamics)
OpenAI outlined a more tempered infrastructure strategy, moving away from an ambitious Nvidia agreement ahead of a potential IPO. (CNBC Tech)
Regulation & Policy
The Trump administration released AI guidelines recommending federal preemption of state laws, innovation focus, and shifting child safety burden to parents. (TechCrunch AI)
The US Defense Department established Palantir’s Maven AI system as a program of record for integration into combat operations. (Bloomberg Technology)
The Department of Defense alleged Anthropic could manipulate models during war, which company executives denied as impossible. (Wired AI)
Democratic senators Warren and Blumenthal queried Nvidia’s $20 billion licensing deal with Groq for potential antitrust violations. (Bloomberg Technology)
Palantir doubled down on AI for battlefield advantage at its developer conference amid business growth. (Wired AI)
Europe shows high AI adoption and talent matching the US but relies on foreign platforms due to infrastructure, regulation, and funding gaps. (The Decoder)
Business & Funding
OpenAI plans to almost double its headcount by end of 2026 to compete with Anthropic and Google. (Bloomberg Technology)
Alibaba and Tencent lost $66 billion in market value after failing to outline clear AI profit visions. (Bloomberg Technology)
xAI is sending engineers to prospective corporate client offices to win business from OpenAI and Anthropic. (Bloomberg Technology)
Uber invested $1.25 billion in Rivian as part of its bet on AI. (AI Business)
AI startup Yotta Data Services seeks $4 billion valuation ahead of IPO, operating India’s largest Nvidia AI processor cluster. (Bloomberg Technology)
Chinese robot maker Unitree Robotics filed for a $610 million IPO in Shanghai. (Bloomberg Technology)
Society & Culture
Nvidia CEO Jensen Huang stated he would be deeply alarmed if a $500K developer spent less than $250K on AI tokens. (The Decoder)
AI tokens are being considered as a potential new pillar of engineering compensation alongside salary. (TechCrunch AI)
Tech workers at companies compete on leaderboards to maximize AI usage, incurring large bills. (NYT Technology)
Companies are using AI as a reason for layoffs amid complex truths. (NYT Technology)
A survey found 95% of UK students use generative AI, with experiences divided between deepening learning and replacing independent thinking. (The Decoder)
Studies show public resistance to AI despite corporate enthusiasm for deployment. (The Verge AI)
FedEx began delivering AI literacy training to over 400,000 workers worldwide. (CNBC Tech)
DoorDash launched a Tasks app where gig workers are paid to record videos of tasks like laundry and eggs to train AI. (Wired AI)
AI tools like Claude are assisting Americans in filing taxes by navigating the US tax system. (Bloomberg Technology)
Crimson Desert developer apologized for using AI-generated art in the game, stating it was intended to be replaced before release. (The Verge AI)
Nvidia’s DLSS 5 AI upscaling technology received criticism from gamers for uncanny visuals. (Wired AI)
Director Valerie Veatch discussed her experience with OpenAI’s Sora text-to-video model and artist communities sharing AI creations. (The Verge AI)
Yale Budget Lab’s Executive Director discussed how 19th-century novels documenting the industrial revolution inform potential societal impacts of the AI revolution. (Bloomberg Technology)