Jason Says
Two things define today: ChatGPT's ad test rewrites the 'free AI' business model at the foundation level, while River AI raising $1.1B at 2 months old proves capital no longer cares about product — only founder pedigree.
OpenAI begins testing ads in ChatGPT to sustain free access, with clear labeling, answer independence, and strong privacy controls. This marks a major business model shift for OpenAI — previously subscription-only — signaling a move toward ad-supported AI, similar to how Google monetized search.
Google's Gemini app has surged to 1 billion users, with 63% using voice interaction daily and over 150 million images generated per day. The data signals a decisive shift toward voice-first AI — Gemini is no longer just a chat tool but a conversational ambient assistant.
OpenAI's longest-serving COO Brad Lightcap is departing to start a new venture, telling staff he's 'excited to advance the mission from a different vantage point.' His exit continues a pattern of senior OpenAI talent spinning out, raising questions about internal stability at the world's most-watched AI company.
OpenAI finally ships a native ChatGPT desktop app for Linux, completing coverage across all three major desktop platforms. This is significant for the developer community — the majority of Linux users are engineers who can now integrate ChatGPT natively into their workflow without a browser.
Abacus AI open-sources Smaug-Agentic, built on Kimi K3 and topping open-source agentic coding leaderboards — just below Claude Opus 5. Available on HuggingFace today and integrating into Abacus AI's agent product. A strong option for developers needing local-deployable, production-grade coding agents.
Addy Osmani (Google Chrome) open-sources agent-skills — packaging senior engineer workflows, quality gates and best practices into reusable skill modules for AI coding agents. Covers the full dev lifecycle from planning to shipping. This fills a critical gap for Claude Code and Cursor users wanting consistent, production-grade agent behavior.
Omega-S solves LLM catastrophic forgetting with just 3 lines added to any training loop — no previous-task data, Fisher matrix, or old weight copies needed. On Llama-3-8B LoRA fine-tuning, it retains original capabilities while adding under 4% compute overhead. Extremely practical for any developer fine-tuning models on custom data.
Business Arena is the first benchmark where LLM agents must actually run a cross-border e-commerce business end-to-end — inferring opportunities from partial data, allocating capital under uncertainty, and meeting regulatory requirements. Developers finally have a realistic yardstick for how capable (or not) today's frontier agents are at real business tasks.
Investors: General Catalyst
xAI 联合创始人 Igor Babuschkin 创立的个人 Agent 公司，成立仅 2 个月即融资 11 亿美元，是近期最具爆炸性的早期 AI 融资案例。
Investors: Accel
Accel 超额认购 5.5 亿美元印度基金在数周内火速关闭，距上一期仅 19 个月，反映全球 VC 对印度 AI+科技市场的强烈信心。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free