8:03 PM
Today in brief.
Google ships Gemini 4 Argon to trusted cyber defenders
DeepMind announced Gemini 4 Argon, its new frontier model for long-horizon coding, enterprise knowledge work, and cyber defense. Access starts with Fairwind cyber partners while Google runs phased safety rollout before paid API and AI Ultra.
🔬 Science
Runway Praxis-1: open-weight video-to-robot control
Runway announced Praxis-1, an open-weight World Action Model that turns video pretraining into robot control policies across embodiments. Partner testing now; public weights promised in coming months.
🔬 Science
SynthID Bio watermarks AI-designed proteins
DeepMind open-sourced SynthID Bio: watermarks that ride in AI-designed protein sequences and 3D structures, survive wet-lab synthesis, and stay detectable with a key. Published in Nature the same day as Argon.
💼 Business
a16z charts: agents burn 5x tokens and power users pull away
Fresh State of Markets II charts: agents burn nearly 5x the tokens humans do (up 14x since February), the top 1% of companies outspend the top 10% on AI by about 8x, and hyperscaler free cash flow collapsed toward chipmakers.
+15 sources 2:39 PM
💼 Business
Runway Ads: autonomous performance marketing with receipts
Runway opened Runway Ads, an autonomous engine for performance creative. Internal numbers since July: 1,000% more ad volume, 2x ROAS, +34% conversion, -41% cost per subscriber, and +$100M ARR.
3 sources 2:35 PM
🔧 Tech
Perplexity open-sources contextual embeddings; email Computer goes free
Perplexity open-sourced pplx-embed-v2-context-9b-preview, a contextual embedding model that encodes each chunk with the whole document in view and leads turbopuffer's context-bench. Separately, anyone can forward tasks to [email protected] for free for a limited time.
+15 sources
🏛️ Policy
Frontier CEOs sign 4-layer AI self-audit pledge
Google, Anthropic, Meta, OpenAI, xAI, and Nvidia signed Trump's Joint Commitment on Frontier Responsibilities: internal controls, an internal assurance team, independent external auditors, and a board committee. Morally binding, not law.
🔧 Tech
Try today: WebGPU local kernels and $5 mic speech
Two open doors for local AI: fastest WebGPU kernels for browser/local inference landed on Hugging Face, and Oído open-source speech recognition beats Whisper-tiny on a $5 microcontroller.
No stories in this category for this edition.