Jason Says
Hallmark turns 'removing AI slop' into a pluggable design skill—that's exactly what the MCP ecosystem should be building. Meanwhile, Grok 4.5 being a real Haiku replacement signals Anthropic's mid-tier moat is eroding faster than most expected.
Together AI released Hallmark, an open-source design skill for Claude Code, Cursor, and Codex. It runs 57 anti-slop test gates and a self-critique pass before emitting any UI, refusing the on-distribution defaults baked into LLM training. Ships with 20 themes and 4 layout macros.
REA is an open-source MCP server that enables AI agents to reverse engineer binary code. It extends the MCP ecosystem into security research, vulnerability analysis, and CTF domains—a notable example of agents gaining low-level binary introspection capabilities.
OpenAI Codex has grown 10x in six months to 7 million users, adding roughly 1M in a single day. Latent Space notes Claude Code's silence on comparable metrics, sparking debate over which AI coding tool actually leads on user traction.
Developer @bindureddy reports Grok 4.5 is a practical Claude Haiku replacement in complex agentic loops, arguing it challenges Anthropic more than OpenAI does. Combined with Meta's Muse Spark, cheap performant models are now eating into proprietary API margins.
Microsoft CEO Satya Nadella warned companies that relying on proprietary AI lab models poses strategic risks, likening them to Trojan horses. The irony: Microsoft is deeply tied to OpenAI. Observers see this as a nudge toward enterprise model ownership—and potentially Microsoft's own AI stack.
Researcher Nathan Lambert argues open-source AI is facing its most serious viability test ever. With closed models widening the performance gap and commercial pressure mounting, the next six months will determine whether open models can carve out a sustainable, differentiated position.
LightMem-Ego is a lightweight streaming memory system for phones and wearables that continuously captures egocentric video/audio and enables long-term experience retrieval. For developers, it offers a viable on-device persistent memory architecture for personal AI assistants—no cloud required.
Long-Horizon-Terminal-Bench introduces 46 long-horizon terminal tasks across 9 categories with dense reward-based grading—moving beyond pass/fail to capture intermediate progress. Directly relevant for evaluating coding agents like Claude Code and Codex on real multi-hour workflows.
AI视频生成创业公司，估值突破20亿美元，资金用于扩展世界模型能力与全球市场布局，是当前视频生成赛道最大单笔融资之一。
Investors: Robot Ventures、USV
Hermes系列开源Agent模型开发商，15亿美元估值融资谈判中，Robot Ventures领投，开源Agent生态的重要玩家。
Investors: IDG资本、蓝思科技、GGG Group、Redstone VC、华山资本、阿联酋磊石资本
通用人形机器人公司，半年累计融资4亿美元，已启动IPO进程，主攻大小脑融合技术与全球规模化部署。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free