<h2>Must know</h2>
<ul>
<li><strong>Pentagon demands military access to AI models</strong>: The Pentagon is demanding AI companies, including Anthropic, allow military use of their technology for “all lawful purposes,” which includes autonomous weapons and mass surveillance. Anthropic is resisting these demands, particularly concerning autonomous weapons, risking the cancellation of its $200 million contract. This dispute highlights the ethical tightrope AI developers walk when their technology intersects with national security, as reported by multiple sources. <a href="https://app.daily.dev/feed-by-ids?id=VtyOOEppn&amp;id=2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI doubles down on agentic AI with key open-source hire</strong>: OpenAI has hired Peter Steinberger, the creator of the open-source personal AI agent OpenClaw, to lead its next-generation agent development. OpenClaw, which automates tasks like email replies, will remain open source under a foundation with OpenAI’s support. This move signals a significant push into agentic AI, where autonomous agents accomplish multi-step goals, raising questions about security, privacy, and the centralization of control in big tech. <a href="https://app.daily.dev/posts/2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 brings M.2 power management and NFS improvements</strong>: Linux 7.0 introduces a power sequencing driver for PCIe M.2 connectors, enabling Device Tree platforms to manage M.2 interfaces, a capability previously limited to ACPI firmware on x86_64 systems. Additionally, it adds dynamic thread pool sizing for NFS servers, allowing the kernel to automatically adjust threads based on workload for optimized resource utilization and performance. Ubuntu 26.04 LTS is also splitting its large <code>linux-firmware</code> package into 17 vendor-specific sub-packages to reduce update sizes and infrastructure strain. <a href="https://app.daily.dev/feed-by-ids?id=2zVNOvhqh&amp;id=8Lxd8Rnjy&amp;id=fP2szhpJX&amp;id=319goy50e" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>LLM inference speeds get a boost with trade-offs</strong>: Anthropic and OpenAI have announced distinct “fast modes” for their coding models, with Anthropic achieving 2.5x speed by reducing batch sizes at 6x the cost, and OpenAI achieving 15x speed using specialized Cerebras chips but requiring a smaller, less capable model. <a href="https://app.daily.dev/posts/e1zDCJ67S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Claude Code gets Google Search Console integration</strong>: A new Python MCP server, <code>mcp-gsc</code>, integrates Google Search Console with Claude Code, enabling SEO analysis and fixes within a single conversational interface, eliminating the need to switch between dashboards and code editors. <a href="https://app.daily.dev/posts/rHujhRHLJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Small team achieves breakthrough in conversational AI latency</strong>: Gradium, a small team from Kyutai, demonstrated Moshi, the first full-duplex conversational AI model to achieve 160ms latency, operating with significantly fewer parameters than typical text models by leveraging deep domain expertise in audio codecs. <a href="https://app.daily.dev/posts/iRysiiuhz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitTalks turns GitHub repos into podcasts</strong>: GitTalks is a new project that converts GitHub repositories into multi-episode podcast playlists by using AI to analyze repository structure, generate conversational dialogue scripts, and synthesize audio, with 80% of its frontend code generated by GitHub Copilot CLI. <a href="https://app.daily.dev/posts/p6iMMSAYk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gwtar introduces new HTML archival format</strong>: Gwtar, a new HTML archival format, aims to solve the “trilemma” of being static, single-file, and efficient by combining an HTML+JavaScript header with a tarball of assets, allowing gigabyte-sized web pages to be archived into a single, lazy-loading file without special server software. <a href="https://app.daily.dev/posts/t2oeOs48x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pharmacy API vulnerability exposed customer data</strong>: A security researcher discovered critical API vulnerabilities in Dava India Pharmacy’s website, allowing unauthorized super admin accounts with full control over 883 stores and access to nearly 17,000 customer orders, which was responsibly disclosed and fixed. <a href="https://app.daily.dev/posts/yhJRbeJMO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>US considers easing China tech restrictions</strong>: The US government is considering reversing some tech restrictions on Chinese companies, potentially removing Alibaba, Baidu, and BYD from military-linked lists, while India has introduced strict new regulations requiring social media platforms to remove AI-generated deepfakes within two hours. <a href="https://app.daily.dev/posts/d9SXCk8xt" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Developers get “Deep Blue” anxiety over AI</strong>: Apparently, software developers are experiencing “Deep Blue” anxiety because AI coding agents can now produce working, tested software autonomously. The article reassures us that accumulated developer experience is still valuable, because, you know, someone still has to tell the AI what to do, right? It’s not like the AI is going to start writing its own requirements documents. <a href="https://app.daily.dev/posts/DwIZWcSFk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding assistants will make us dumb</strong>: An opinion piece warns that over-reliance on AI coding assistants might lead to “cognitive atrophy” in programmers, suggesting we’ll lose our problem-solving skills if we don’t actively practice coding logic. Because, clearly, the only way to keep our brains sharp is to manually write every line of code, rather than, say, focusing on higher-level architectural challenges or actual problem-solving that AI can’t do. <a href="https://app.daily.dev/posts/qtVi65Njq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Alibaba’s new AI image model struggles with basic tasks</strong>: Alibaba released Quinn Image 2.0, an AI image generation model, claiming improved typography and 2K native resolution. However, one user’s attempt to generate a YouTube thumbnail resulted in issues with aspect ratio and output quality, suggesting that despite the hype, these models still struggle with basic instructions and practical application. Maybe it’s not the AI that doesn’t understand the assignment, but the marketing department. <a href="https://app.daily.dev/posts/zRxAENzbc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
