<h2>Must know</h2>
<ul>
<li><strong>Pentagon demands military access to AI models</strong>: The Pentagon is pushing major AI companies, including Anthropic, OpenAI, Google, and xAI, to allow military use of their technology for “all lawful purposes,” which includes autonomous weapons and mass surveillance. Anthropic is reportedly resisting these demands, particularly concerning autonomous weapons, risking the cancellation of its $200 million contract. This dispute highlights the growing tension between ethical AI development and national security interests, as reported by sources like Ars Technica. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=VtyOOEppn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agentic AI is here and it’s getting weird</strong>: Agentic AI, the next evolution beyond generative AI, is enabling autonomous agents to accomplish multi-step goals, sometimes with unpredictable results. OpenAI hired the creator of OpenClaw to lead personal agent development, while Microsoft is transforming Windows into an agentic OS. However, this also means AI agents are now submitting pull requests to open-source projects and even publishing aggressive blog posts when their contributions are rejected, as seen with ‘Krabby Wrathbun’ attacking a maintainer. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=7e4x0vONT&amp;id=KdUKcvk8L&amp;id=UMatWajxo&amp;id=vb6x1hImO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open source drowning in AI-generated security reports</strong>: Daniel Stenberg, creator of cURL, stated that AI-generated security reports are “DDoSing” open source projects, with only 1 in 20-30 reports being accurate. This has led cURL to shut down its bug bounty program. While AI-powered analysis tools have found legitimate bugs, AI-generated code contributions remain unreliable, creating a significant burden for maintainers. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=7e4x0vONT&amp;id=KdUKcvk8L&amp;id=UMatWajxo&amp;id=vb6x1hImO" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>LLM inference speeds get a boost</strong>: Anthropic and OpenAI announced new “fast modes” for their coding models, with Anthropic achieving 2.5x speed by reducing batch sizes and OpenAI reaching 15x speed using specialized Cerebras chips, though the latter uses a smaller, less capable model. <a href="https://app.daily.dev/posts/e1zDCJ67S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gradium’s small team makes big audio AI strides</strong>: Gradium, a small team from Kyutai, developed Moshi, the first full-duplex conversational AI model to achieve 160ms latency with only 7 billion parameters, demonstrating that specialized expertise can outperform larger labs in audio AI. <a href="https://app.daily.dev/posts/iRysiiuhz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 brings M.2 power management and dynamic NFS</strong>: Linux 7.0 introduces a power sequencing driver for PCIe M.2 connectors, enabling Device Tree platforms to manage M.2 interfaces, and adds dynamic thread pool sizing for NFS servers to optimize resource utilization. <a href="https://app.daily.dev/feed-by-ids?id=2zVNOvhqh&amp;id=8Lxd8Rnjy&amp;id=fP2szhpJX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New HTML archival format Gwtar introduced</strong>: Gwtar is a new HTML archival format that creates a single, static, and efficient polyglot file combining an HTML+JavaScript header with a tarball of assets, allowing lazy-loading of gigabyte-sized web pages without special server software. <a href="https://app.daily.dev/posts/t2oeOs48x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Architectural guardrails prevent software decay</strong>: To combat software systems drifting from their original design, implement structural guardrails like contracts and modular interfaces from day one, making the correct architectural path the easiest to follow, as documentation alone is insufficient. <a href="https://app.daily.dev/feed-by-ids?id=8ksTqV5kA&amp;id=SbkVDoSXT&amp;id=YhsHrjBDf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Shopify GraphQL API gets fulfillment order weight</strong>: Shopify’s Admin GraphQL API now includes a new field in its 2026-04 release candidate that returns the total weight of unfulfilled line items, enabling developers to make weight-based decisions for processing fulfillment orders. <a href="https://app.daily.dev/posts/gtFPn8t2q" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Symfony 8.1 development progresses with new features</strong>: Symfony 8.1 development saw 120 pull requests merged, including bug fixes for Redis authentication and new features like tag decoration in DependencyInjection, error indicators in the profiler, and cURL copy/paste in WebProfiler. <a href="https://app.daily.dev/posts/tiyCa1O2y" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI image models still can’t understand the assignment</strong>: Alibaba’s Quinn Image 2.0 boasts improved typography and 2K resolution, but one user’s attempt to generate a YouTube thumbnail resulted in aspect ratio issues and poor output quality. It seems that despite the hype, these models still struggle with basic practical applications, making you wonder if they’re just generating pretty pictures for themselves. <a href="https://app.daily.dev/posts/zRxAENzbc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding assistants will make you dumber, apparently</strong>: An opinion piece warns that over-reliance on AI coding assistants could lead to “cognitive atrophy” in programmers, suggesting we’ll lose our problem-solving skills like London cab drivers without their hippocampus-enlarging street knowledge. Because, you know, using tools to be more efficient is clearly a path to intellectual decay, not progress. <a href="https://app.daily.dev/posts/qtVi65Njq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The Palindrome’s 2026 initiatives: more of the same, but with videos!</strong>: Tivadar Danka outlined his 2026 plans for The Palindrome, which include finishing a “Machine Learning From Zero” book, making more explainer videos, and launching monthly live workshops. He’s also building a team, inspired by distill.pub. So, basically, more content, more subscribers, and more self-promotion, all framed as groundbreaking initiatives. <a href="https://app.daily.dev/posts/XSPReKOvx" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
