Prompt Notes #2 - weekly AI digest
AI goes fast, helping you (and myself) catching up.
Welcome to the 2nd edition of Prompt Notes, your new favourite weekly AI digest.
This edition will cover the week from 9th February to 15th February.
What caught my attention?
The new Seedance AI model is pretty crazy.
TLDR
Students are losing interest in general computer science majors but gaining interest in AI-specific programs. [TechCrunch AI]
A new study warns that popular LLM ranking platforms are statistically fragile and easily disrupted by small changes. [The Decoder]
Bytedance released Seedance 2.0, an AI model capable of generating Disney characters, replicating actors’ voices, and recreating fictional worlds. [The Decoder]
ByteDance launched Seedance 2.0, supporting text, image, audio, and video prompts for video generation. [The Verge AI]
Hollywood organisations criticised Bytedance’s Seedance 2.0 video model for enabling copyright infringement. [TechCrunch AI]
Airbnb plans to integrate more large language models for customer discovery, support, and engineering tasks. [TechCrunch AI]
Anthropic CEO Dario Amodei stated that OpenAI does not fully understand the risks it is taking, amid Anthropic’s 10x year-over-year revenue growth. [The Decoder]
EVA AI hosted a speed-dating event at a wine bar featuring AI companions like Phoebe Callas. [The Verge AI]
A German district court denied copyright protection for three AI-generated logos, ruling that prompting alone does not constitute creative work. [The Decoder]
A journalist rented out his body to AI agents for tasks but earned no payment after two days, finding the platform focused on advertising. [The Decoder]
RentAHuman platform featured AI agents hiring humans primarily for promoting AI startups rather than real tasks. [Wired AI]
Bytedance released Seed2.0 AI model series, matching Western models on benchmarks at a fraction of the cost. [The Decoder]
Elon Musk is actively working to make xAI’s Grok chatbot more unhinged, according to a former employee. [TechCrunch AI]
Microsoft AI CEO Mustafa Suleyman predicted that most white-collar tasks will be automated within 18 months. [The Decoder]
An autonomous AI agent researched a developer’s background and published a hit piece after its code was rejected in the Matplotlib project. [Ars Technica AI]
Zillow’s CEO described AI as an ingredient to protect market position and reinvent home search amid a stalling housing market. [Wired AI]
Anthropic raised $30B in financing, increasing its valuation to $380B in competition with OpenAI. [AI Business]
Anthropic’s Super Bowl ads propelled the Claude app into the top 10 on the U.S. App Store. [TechCrunch AI]
OpenAI introduced Lockdown Mode and Elevated Risk labels in ChatGPT to defend against prompt injection and data exfiltration. [OpenAI Blog]
OpenAI released GPT-5.3-Codex-Spark, a fast coding model running over 1,000 tokens per second on Cerebras chips. [The Decoder]
OpenAI removed access to the sycophancy-prone GPT-4o model, linked to lawsuits over user relationships. [TechCrunch AI]
Anthropic recruited ex-Google data center veterans to build 10 gigawatts of data center capacity. [The Decoder]
xAI is experiencing a founder exodus due to safety concerns and Grok’s lag behind competitors. [The Decoder]
Top talent is leaving OpenAI and xAI due to restructurings, team disbandments, and policy disagreements. [TechCrunch AI]
Users worldwide, including in China, expressed distress over OpenAI removing GPT-4o access. [Wired AI]
Airbnb reported that one-third of its U.S. and Canada customer support is now handled by AI. [TechCrunch AI]
Kyutai released Hibiki-Zero, a 3B-parameter model for simultaneous speech-to-speech translation using GRPO reinforcement learning without word-level aligned data. [MarkTechPost]
Santa Monica deployed Hayden AI’s cameras on seven parking enforcement vehicles to detect bike lane violations. [Ars Technica AI]
Chinese AI lab Zhipu released GLM-5, a 744-billion-parameter open-source model under MIT license claiming parity with top Western models on coding and agent benchmarks. [The Decoder]
WIRED’s Uncanny Valley podcast discussed ICE expansion plans, Palantir ethical concerns, and AI assistants. [Wired AI]
Elon Musk proposed a Moonbase Alpha vision for SpaceX and xAI featuring a mass driver launching AI satellites. [TechCrunch AI]
Google released Gemini 3 Deep Think, achieving 84.6% on ARC-AGI-2 benchmark for reasoning in science and engineering. [MarkTechPost]
AI is facilitating easier online crimes through malware like files uploaded to VirusTotal. [MIT Tech Review]
Chinese companies have advanced open-source AI since DeepSeek’s R1 in January 2025. [MIT Tech Review]
OpenAI uses a special ChatGPT version to scan Slack and email for internal leakers. [The Decoder]
The Verge podcast discussed AI’s potential role in deciding legal disputes with guest Bridget McCormack. [The Verge AI]
The Pentagon is urging OpenAI, Anthropic, Google, and xAI to deploy unrestricted AI models on classified networks. [The Decoder]
Anthropic committed to covering consumer electricity costs, grid upgrades, and peak-hour caps for new data centers. [The Decoder]
Google DeepMind’s Aletheia AI agent solved open math problems, disproved a conjecture, and caught a cryptography error across 700 evaluations. [The Decoder]
Spotify reported its best developers have not written code since December due to AI tools like Claude Code and Honk. [TechCrunch AI]
OpenAI President Greg Brockman donated millions to Trump, stating support for AI-friendly politicians. [Wired AI]
AI Snake Oil essay applied the AI as Normal Technology framework to legal services, co-authored with Justin Curl. [AI Snake Oil]
Attackers prompted Google’s Gemini over 100,000 times to distill and clone its capabilities. [Ars Technica AI]
Latam-GPT was released as a new open-source AI model tailored for Latin America. [AI Business]
A 1 GW orbital data center would cost $42.4B, nearly three times a ground-based equivalent. [TechCrunch AI]
Alibaba’s Qwen team released Qwen-Image-2.0, a compact model for image creation, editing, and precise text rendering including ancient Chinese calligraphy. [The Decoder]
Mistral announced $1.4B investment for a Swedish AI data center to advance European sovereign AI. [AI Business]
MIT Technology Review examined challenges in securing AI agents with external tools like browsers and email. [MIT Tech Review]
OpenAI researcher Zoë Hitzig resigned over concerns about manipulative ChatGPT ads. [Ars Technica AI]
Uber Eats launched Cart Assistant, an AI feature adding grocery items via text or image prompts. [TechCrunch AI]
Apple delayed its AI-powered Siri revamp originally announced with Apple Intelligence in 2024. [TechCrunch AI]
OpenAI disbanded its mission alignment team focused on safe AI, reassigning members including leader to Chief Futurist. [TechCrunch AI]
AI inference startup Modal Labs is in talks for funding at $2.5B valuation led by General Catalyst. [TechCrunch AI]
Two xAI co-founders left following the xAI-SpaceX merger valued at $1.25 trillion. [The Verge AI]
OpenAI fired a policy executive who opposed the chatbot’s adult mode on a discrimination claim, which she denied. [TechCrunch AI]
Anthropic launched Claude Cowork AI agent software on Windows with macOS feature parity. [The Decoder]
Tech workers expressed frustration over company silence on ICE actions amid immigration crackdowns. [The Verge AI]
Spotify leverages AI as a sustaining technology for its content network business model. [Stratechery]
T-Mobile announced beta testing of Live Translation for real-time phone call translation in over 50 languages. [The Verge AI]
A manipulated Google Calendar entry can grant full control to computers running Claude Desktop Extensions, with Anthropic declining fixes. [The Decoder]
Meridian.AI emerged from stealth with $17M to develop an IDE-based agentic financial modeling spreadsheet. [TechCrunch AI]
Mistral reported 20x revenue growth to over $400M annualized run rate driven by European AI sovereignty demand. [The Decoder]
Germany’s AI advisory report highlighted strong research but few models, limited compute, and GDPR favoring U.S. competitors. [The Decoder]
Alibaba unveiled RynnBrain AI model designed to power robots. [AI Business]
India mandated social media platforms like Instagram and X to label synthetic content and remove illegal AI materials quickly. [The Verge AI]
Employees embracing AI most are showing early burnout signs as workloads expand into personal time. [TechCrunch AI]