Prompt Notes #1 - weekly AI digest
AI goes fast, helping you (and myself) catching up.
Welcome to the 1st edition of Prompt Notes, your new favourite weekly AI digest.
The promise is simple: 5 minutes for you to catch up on what happened in AI, with clickable links. Alongside that, I’ll share something that piqued my interest over that week.
Note: the image is AI-generated, unlike the Patch Notes one, and I also use AI tools in my workflow to sort out the news.
This edition will cover the week from 2nd February to 8th February.
What caught my attention?
YCombinator’s interview with OpenClaw’s founder. YC is still highly influential in the tech world, so having them chat with the creator of this new platform is telling.
Here’s the official description of the interview:
”You’ve probably already heard all about OpenClaw (formerly Clawdbot/Moltbot). The viral sensation is an open-source AI assistant that runs on your own device, connects with messaging apps you already use, and goes beyond chat to actually execute tasks like managing your email, calendars, files, workflows, and more. Now meet the man behind it. YC’s Raphael Schaad sat down with Peter Steinberger, the creator of OpenClaw, to discuss the “aha” moment behind the viral personal AI agent, why local-first agents could replace many of today’s apps, and how personal agents will reshape the future of software.”
A TLDR of the video:
OpenClaw’s Viral Success (0:00): The GitHub repo for OpenClaw (formerly Clawdbot/Moltbot) gained over 160,000 stars rapidly. The community has developed numerous projects, including Moltbook, in which bots communicate with one another and even hire humans for real-world tasks.
The “Aha” Moment (5:07): Peter describes his journey to creating OpenClaw, from his initial desire for a computer that could execute typed commands to the pivotal moment when the agent autonomously performed complex tasks beyond his anticipation, such as converting a voice message to text by finding and using an external API.
Local-First AI Agents (1:28): Peter emphasises that OpenClaw’s ability to run directly on a user’s computer, rather than in the cloud, is its key differentiator, allowing it to interact with and control various aspects of the user’s digital environment.
The Future of Apps (10:21): Peter speculates that 80% of current applications, particularly those focused on data management (e.g., fitness trackers, to-do lists), will disappear as personal AI agents take over these functions.
Data Ownership and Privacy (12:31): The conversation highlights the importance of data ownership, noting that OpenClaw stores user memories as Markdown files directly on their machines, ensuring privacy and control over sensitive personal information.
Contrarian Building Philosophy (18:19): Peter shares his unconventional approach to software development, favouring command-line interfaces (CLIs) over multi-command processors (MCPs) and emphasising simplicity and directness in his workflow.
TLDR
Employees embracing AI report early signs of burnout as productivity gains lead to expanded to-do lists encroaching on lunch breaks and evenings. TechCrunch AI
Top multimodal models, including Gemini 3 Pro at 47.4%, fail to exceed 50% accuracy on the WorldVQA benchmark for specific visual entity recognition, such as species or product names. The Decoder
Claude Opus 4.6 ranked first on the Artificial Analysis Intelligence Index, surpassing GPT-5.2, though OpenAI’s Codex 5.3 is upcoming and has lower token costs. The Decoder
Research shows reasoning models like Deepseek-R1 simulate internal “societies of thought” with diverse expert personas, including extraverted and neurotic traits, improving problem-solving performance. The Decoder
The New York legislature is considering two bills: the NY FAIR News Act, which would require labels on AI-generated news content, and a three-year pause on new data centre construction. The Verge AI
Anthropic launched Claude Fast Mode for Opus 4.6, delivering 2.5x faster responses at matching quality but up to 6x higher token prices. The Decoder
ByteDance released Protenix-v1, an open-source model under Apache 2.0, achieving AlphaFold3-level performance in biomolecular structure prediction for proteins, DNA, RNA, and ligands. MarkTechPost
AI-generated fake videos proliferated on Japanese social media during the lower house election, with over half of the surveyed respondents believing the misinformation. The Decoder
Google AI and Peking University introduced PaperBanana, a multi-agent framework automating publication-ready methodology diagrams and statistical plots for research workflows. MarkTechPost
AI-powered biotech startups at Web Summit Qatar use automation, data analysis, and gene editing to address labour shortages in rare disease drug discovery and treatment. TechCrunch AI
Apple scaled back its AI-powered virtual health coach project codenamed Mulberry amid new leadership demands for faster results. The Decoder
OpenAI CEO Sam Altman stated AI agents will autonomously write code to integrate services, regardless of official APIs. The Decoder
Moltbook launched as a Reddit-like social network for AI agents to share, discuss, and upvote content, with humans observing. MIT Tech Review
Waymo introduced the Waymo World Model, a generative simulator built on Genie 3 for photorealistic, controllable multi-sensor autonomous driving scenes. MarkTechPost
Apple is developing CarPlay support for third-party voice control apps from OpenAI, Anthropic, Google, and others, bypassing iPhone access. The Verge AI
WordPress now integrates Claude for analysing web traffic and internal site metrics. TechCrunch AI
A New York federal judge dismissed a case for a lawyer’s repeated misuse of AI, including fake citations and irrelevant references to Ray Bradbury. Ars Technica AI
Improved AI measurement could unlock $32B for marketers by addressing biases and underfunding in unmeasured channels. MarTech
OpenAI CEO Sam Altman claimed ChatGPT has more users in Texas alone than Claude has across the entire US. The Decoder
Mistral AI launched Voxtral Transcribe 2, featuring batch diarization and open, real-time ASR models optimised for multilingual production workloads. MarkTechPost
NVIDIA released VibeTensor, an open-source deep learning runtime generated end-to-end by LLM-powered coding agents spanning Python, JavaScript, C++, and CUDA. MarkTechPost
Meta completed preliminary training of its new AI model codenamed Avocado, per an internal memo. The Decoder
Google Cloud developed an AI tool for U.S. Ski and Snowboard to analyse and elevate athletes’ tricks ahead of the 2026 Winter Olympics. Google AI Blog
Anthropic’s Claude and Claude Code gained significant popularity, with its creator, Boris Cherny, recognised publicly. The Verge AI
X is testing collaborative Community Notes where AI composes initial notes for human refinement. Social Media Today
OpenAI launched Frontier, an enterprise platform for building, deploying, and managing AI agents with shared context, onboarding, permissions, and governance. OpenAI Blog
Bundesliga club VfL Wolfsburg integrated ChatGPT club-wide to enhance efficiency, creativity, and knowledge while preserving its identity. OpenAI Blog
NVIDIA is reportedly investing $20 billion in OpenAI’s next funding round, proceeding cautiously amid downplayed tensions. The Decoder
Anthropic announced partnerships with major US research institutions to develop AI agents addressing biology’s data bottlenecks. The Decoder
Salesforce executive emphasised data foundations with context as essential for advancing from LLMs to AI agents. MarTech
ElevenLabs raised 500 M from Sequoia Capital at an 11 billion valuation, tripling its value in 12 months. TechCrunch AI
Mistral released an ultra-fast translation model, Voxtral, that competes with major US AI labs by efficiently using GPUs. Wired AI
Kling AI launched version 3.0 with support for longer clips, improved character consistency, and 4K image generation. The Decoder
Tinder began testing AI-powered recommendations and Camera Roll insights to improve matches and combat swipe fatigue. TechCrunch AI
AI math startup Axiom solved four previously unsolved math problems using its advanced reasoning AI. Wired AI
Andreessen Horowitz raised $1.7B for AI infrastructure investments in companies like Black Forest Labs, Cursor, and OpenAI. TechCrunch AI
Anthropic’s AI plugins have raised concerns in the legal industry about competition with domain-specific tools and potential impacts on workers. AI Business
Alphabet reported over $400 billion in annual revenue for the first time in its Q4 2025 earnings, driven by cloud and YouTube growth. The Verge AI
Google’s Gemini app reached over 750 million monthly active users. TechCrunch AI
Google introduced Agentic Vision in Gemini 3 Flash, enabling active, tool-using loops for detailed image understanding beyond single-pass processing. MarkTechPost
If you liked this, don’t hesitate to share!
Until next time,
Rachid