Prompt Notes #6 - Weekly AI Digest
AI goes fast. Helping you (and myself) catch up.
Welcome to the 6th edition of Prompt Notes, your new favourite weekly AI digest.
This edition will cover the week from 9 March to 15 March.
What caught my attention
Elon Musk built xAI fast and loud, and now he’s admitting it wasn’t built right. The restructuring, the co-founders’ exits, and now the poaching from Cursor - it all traces back to the same roots: “move fast and break things” (a Silicon Valley favourite) and… ego.
This is fine when the stakes are low. But when you have billions of dollars engaged in computing, a war with OpenAI and Google, and a defence sector watching closely - that’s a whole other ball game.
TLDR
xAI full restructuring — Musk admitting the company wasn’t built right, co-founders gone, and Cursor execs brought in signals deeper structural rot than a typical pivot. (The Decoder)
Anthropic vs. the Pentagon Electric Boogaloo Part 2 — The US War Department calling Claude an ethical liability while Palantir keeps using it anyway is the most revealing AI/defense tension of the year so far. (The Decoder)
ChatGPT is losing ground fast — Market share dropped from 75.7% to 61.7% while Gemini went from 5.7% to 24.4%. The “default AI” position is no longer locked in. (The Decoder)
China’s OpenClaw surge — Usage exceeding the US, government subsidies for one-person companies, Alibaba deploying it in minutes. The agentic race has a geography now. (CNBC Tech)
Cursor at $50 billion — An AI coding tool in talks for a $50B valuation while xAI is raiding it for talent. Ironic. (Bloomberg Technology)
Rest of the news
Models & Research
Princeton’s OpenClaw-RL framework converts live signals from chats, terminal commands, and GUI actions into continuous training data for AI agents, enabling improvements from a few dozen interactions. (The Decoder)
Google DeepMind introduced Aletheia, an AI agent for autonomous professional research that iteratively generates, verifies, and revises solutions. (MarkTechPost)
xAI’s Grok 4.20 set a new hallucination record low but trailed Gemini and GPT-5.4 in benchmarks by a wide margin. (The Decoder)
Meta postponed its AI model Avocado after internal tests showed it lagging behind Google, OpenAI, and Anthropic. (The Decoder)
Former Anthropic researchers launched Mirendil, an AI startup targeting scientific research in biology and materials science, seeking a $1 billion valuation. (The Decoder)
Model Context Protocol (MCP) and AI Agent Skills compared across setup, task execution, and behavioral guidance for LLMs interacting with tools. (MarkTechPost)
Analysis of the current AI phase and likely next developments in human-AI work by Ethan Mollick. Note: highly recommend his book “Co-Intelligence”. (One Useful Thing)
Products & Tools
Anthropic launched a beta feature for Claude to generate interactive charts, diagrams, and visualizations inline in conversations. (The Decoder)
Perplexity launched Personal Computer, a $200/month AI agent handling emails, presentations, and app control around the clock. (The Decoder)
Microsoft launched Copilot Health, an AI assistant integrating wearables, medical records, and lab results for personalized health advice. (The Decoder)
Google launched Ask Maps in Google Maps, using Gemini for plain-language place searches and revamped 3D navigation. (The Decoder)
X enabled Grok users to generate video incorporating up to seven input images. (Social Media Today)
LangChain released Deep Agents, a structured runtime for planning, memory, and context isolation in multi-step AI agents. (MarkTechPost)
Garry Tan released gstack, an open-source Claude Code system with eight workflow skills for planning, code review, QA, and shipping, backed by persistent browser runtime. (MarkTechPost)
OpenAI plans to integrate Sora video AI into ChatGPT. (The Decoder)
Bumble announced AI dating assistant Bee for compatibility-based matching beyond swiping. (TechCrunch AI)
Genspark launched Claw AI assistant using dedicated cloud computers as a secure OpenClaw alternative. (SiliconANGLE AI)
Google released a breakdown of differences between its three Nano Banana image generation models, with Nano Banana 2 offering 95% of Pro capabilities plus web search. (The Decoder)
Google and Samsung rolled out Gemini task automation on S26 and Pixel 10 for apps including food delivery and rideshare. (The Verge AI)
Alibaba launched a mobile app to install and deploy OpenClaw within minutes. (Bloomberg Technology)
LinkedIn published a guide on optimizing post content for AI chatbots to improve visibility in AI-generated answers. (Social Media Today)
Bespoke AI models are emerging for filmmaking, moving beyond general image and video generators. (The Verge AI)
Google uses old news reports and LLMs to convert qualitative data into quantitative inputs for flash flood prediction. (TechCrunch AI)
Infrastructure & Compute
Nvidia and Nebius partnered to develop a hyperscale AI cloud, with Nvidia investing $2 billion for 5GW deployment by 2030. (Capacity Global)
AWS partnered with Cerebras for AI inference disaggregation to be deployed on Amazon Bedrock. (Data Center Dynamics)
Meta unveiled four generations of custom AI chips focused on inference to reduce GPU dependence. (The Decoder)
xAI received approval for 41 natural gas turbines in Mississippi to power its Colossus data centers. (Data Center Dynamics)
Hyperscale Power raised €5M seed to build a prototype miniaturized solid-state transformer for data center power. (Data Center Dynamics)
Nvidia plans $26 billion investment in open-weight AI models over five years, per SEC filing. (The Decoder)
Regulation & Policy
US War Department CTO claimed Anthropic’s Claude “pollutes” the defense supply chain with built-in ethics. (The Decoder)
Microsoft, OpenAI and Google employees, ex-military, and rights groups filed amicus briefs backing Anthropic in its legal battle against the Pentagon. (The Decoder)
A lawyer linked AI chatbots to mass casualty cases and warned of risks as the technology advances faster than safeguards. (TechCrunch AI)
Palantir demos show Anthropic’s Claude helping the Pentagon analyze intelligence and suggest next steps for war plans. (Wired AI)
Sam Altman faced questions from lawmakers on OpenAI’s Defense Department work. (CNBC Tech)
Google is not ruling out ads in Gemini. (Wired AI)
Google’s generative AI search tools increasingly cite its own services like Google Search and YouTube over third-party publishers. (Wired AI)
Business & Funding
xAI is revamping its AI coding tool with two new executives hired from Cursor. (TechCrunch AI)
Andreessen Horowitz’s Top 100 AI ranking shows ChatGPT leading a maturing market with growing competitors and geopolitical usage splits. (The Decoder)
Similarweb data shows ChatGPT market share fell from 75.7% to 61.7% as Google Gemini rose from 5.7% to 24.4%. (The Decoder)
Anthropic removed the surcharge for million-token context windows in Claude Opus 4.6 and Sonnet 4.6, reducing costs for requests over 200,000 tokens. (The Decoder)
Anthropic launched the Anthropic Institute to study AI risks, led by co-founder Jack Clark. (SiliconANGLE AI)
AI coding startup Cursor is in talks for a $50 billion valuation funding round. (Bloomberg Technology)
Nick Clegg launched AI startup Efekta, deliberately avoiding an AGI focus after his time at Meta. (Wired AI)
Elon Musk’s ketamine use ruled off-limits in OpenAI fraud trial. (Bloomberg Technology)
Microsoft is pushing AI adoption in Africa, competing with DeepSeek. (Bloomberg Technology)
Society & Culture
ByteDance shelved the global launch of Seedance 2.0 after copyright complaints from Hollywood studios. (The Decoder)
An Australian AI consultant used ChatGPT, AlphaFold, and Grok to identify a possible treatment for his dog’s incurable cancer. (The Decoder)
ServiceNow CEO stated AI agents could push college graduate unemployment over 30%. (CNBC Tech)
Seven Chinese local governments launched million-dollar funding programs for OpenClaw projects to enable “one-person companies” with AI agents. (The Decoder)
China’s OpenClaw usage exceeded the US, driving demand for lower-cost Chinese AI models. (CNBC Tech)
Google expanded Gemini in Chrome to India, Canada, and New Zealand with support for Hindi, Bengali, and other languages. (TechCrunch AI)
Until next week!
Rachid