<h2>Must know</h2>
<ul>
<li><strong>Yann LeCun launches AMI Labs to build ‘world models’</strong>: Yann LeCun, a prominent figure in AI, has launched AMI Labs, a new startup focused on developing “world models” for intelligent systems. The company aims to offer an alternative to large language models (LLMs) by emphasizing persistent memory, reasoning, and safety. AMI Labs, co-founded with Alex LeBrun, is reportedly raising funding at a $3.5 billion valuation and plans to target high-stakes sectors like healthcare and robotics, as reported by various sources covering the announcement. <a href="https://app.daily.dev/posts/CvMr0F9gc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Oracle sorting algorithm patent expires, open source benefits</strong>: Oracle’s patent for the “Orasort” sorting algorithm has expired after 20 years, opening the door for its adoption in open-source databases like MySQL and PostgreSQL. The algorithm, which combines common prefix skipping, adaptive quicksort/radix sort, and key substring caching, claims to offer up to a 5x performance improvement. Mark Callaghan, the inventor, confirmed the details, sparking immediate interest in the open-source community for integration. <a href="https://app.daily.dev/posts/aheyA2T6F" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TikTok’s new privacy policy expands data collection</strong>: TikTok’s updated U.S. privacy policy now explicitly includes precise GPS location tracking for U.S. users, tracking of AI interactions (prompts and outputs), and expanded use of ad network data for targeted advertising across the web. Users must accept these terms to continue using the app. While concerns about “immigration status” collection were raised, privacy lawyers clarified this is a legal disclosure requirement, not a new collection practice, as reported by TechCrunch and other outlets. <a href="https://app.daily.dev/feed-by-ids?id=BFPMM0rX7&amp;id=nbFKCoxnX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Newelle AI assistant for GNOME adds local LLM support</strong>: Newelle, the AI assistant for the GNOME desktop, now supports local large language models (LLMs) via llama.cpp and can execute system commands, expanding its utility beyond cloud AI APIs like Google Gemini and OpenAI. <a href="https://app.daily.dev/posts/Au9UNs8LJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust-based deep learning framework Burn released</strong>: Burn is a new deep learning framework and tensor library written in Rust, offering flexibility, efficiency, and portability across various hardware backends including CUDA, ROCm, and Metal, aiming to bridge the Python/C++ gap in ML. <a href="https://app.daily.dev/posts/EAj5GDC5K" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLM observability challenges traditional monitoring</strong>: Traditional observability methods are falling short for LLM-powered applications due to their probabilistic and multi-step nature, necessitating new approaches like tracking prompt versions, agent pipelines, and token usage, with eBPF emerging as a key infrastructure-level solution. <a href="https://app.daily.dev/posts/2NT8yV81n" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 6.19 kernel adds ASUS Armoury driver</strong>: The Linux 6.19 kernel now includes the ASUS Armoury driver, enhancing support for ASUS ROG gaming devices, including ROG Ally handhelds and various ROG laptops. <a href="https://app.daily.dev/posts/rvv41Cn0C" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rails updates bring performance and maintenance improvements</strong>: Rails introduced a 2x speed improvement to <code>String#squish</code>, enhanced error handling for stream rendering, and improved the <code>console reload!</code> command, alongside better generator detection for package managers and optimized <code>insert_all!</code> for composite primary keys. <a href="https://app.daily.dev/feed-by-ids?id=WT6Ixi3bR&amp;id=Ji6vBmphI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GNU C Library 2.43 enhances C23 support and system calls</strong>: GNU C Library 2.43 expands C23 language support, adds new system call wrappers for Linux (mseal, openat2), improves FMA performance on AMD Zen, and enables 2MB transparent hugepages for malloc on AArch64. <a href="https://app.daily.dev/posts/4kkaXLeFv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wix releases React Native UI component library</strong>: Wix has launched React Native UI Lib, a new UI component library offering pre-built, customizable components with integrated theme management and modifier-based styling for React Native applications. <a href="https://app.daily.dev/posts/3XpEcphch" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding tools: the new ‘hidden’ feature that makes AI a ‘team lead’</strong>: Apparently, Claude Code has a ‘hidden’ feature called ‘Swarms’ that elevates the AI from a mere coder to a ‘team lead’ role, generating multiple specialist agents to coordinate tasks. This ‘undocumented’ capability, discovered by some developer, sounds suspiciously like a fancy way to describe a more complex prompt engineering setup, rather than a revolutionary AI breakthrough. It’s almost as if they’re trying to make us believe the AI is managing itself, instead of just executing a more elaborate script we wrote. <a href="https://app.daily.dev/feed-by-ids?id=BCMHZegjN&amp;id=LMNhcIFC9&amp;id=kVmMac2gZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Software estimation: still a political game, not engineering</strong>: According to some ‘experts,’ software estimation is ‘fundamentally impossible’ because of all the ‘unknown unknowns,’ and managers just use estimates as ‘political tools.’ So, after decades of software development, we’re still pretending that engineering is about guessing and negotiating, rather than, you know, engineering. Maybe if we stopped calling it ‘estimation’ and started calling it ‘wild speculation with a side of corporate theater,’ we’d all be more honest. <a href="https://app.daily.dev/feed-by-ids?id=HxseyW07U&amp;id=vTe4dHYnt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered educational app for kids: because what could go wrong?</strong>: Sparkli, a startup by ex-Google employees, just secured $5 million to build an AI-driven interactive learning app for kids aged 5-12, generating multimedia content on-demand in two minutes. They’re piloting it in schools and touting ‘built-in safety features.’ Because nothing says ‘safe and educational’ like AI-generated content for children, especially when there’s no independent assessment of its educational value or safety. What could possibly go wrong with an algorithm teaching financial literacy to a 5-year-old? <a href="https://app.daily.dev/posts/CoMIJ1Dxu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
