<h2>Must know</h2>
<ul>
<li>New AI models push boundaries, China challenges US chip dominance: NVIDIA released Llama Nemotron Super v1.5, a 49 billion parameter model optimized for reasoning, outperforming larger models and achieving 3x throughput. Huawei countered with CloudMatrix 384, an AI computing system with 384 of its 910C chips, directly targeting Nvidia’s GB200 NVL72 system. China’s Premier Li Qiang announced plans for an international AI cooperation organization to resist monopolistic control, implicitly referencing US export restrictions on AI chips, as reported by TechCentral. This intensifies the global AI leadership race and hardware rivalry.</li>
<li>Compact AI models show promise for reasoning: The Hierarchical Reasoning Model (HRM), a new recurrent neural network, achieves state-of-the-art reasoning with only 27 million parameters. It uses interdependent planning and computation modules, outperforming larger models on complex tasks like Sudoku and the Abstraction and Reasoning Corpus benchmark. This suggests that architectural advances can yield powerful AI without massive parameter scaling, challenging the current trend.</li>
<li>Higgsfield framework simplifies large-scale AI training: Higgsfield is an open-source machine learning framework designed for scalable, fault-tolerant GPU orchestration in massive neural network training. It simplifies distributed training complexity and integrates with PyTorch and DeepSpeed APIs. This framework could democratize access to ultra-large model research for teams with diverse infrastructure, as detailed on its GitHub repository.</li>
<li>AI transforms workplaces, raising concerns about ‘cognitive migration’: AI adoption is rapidly changing workplaces, leading to a concept of ‘cognitive migration’ that divides labor into five groups: willing adopters, pressured migrants, resistant workers, unreached workers, and disconnected individuals, according to VentureBeat. Projections suggest potential 10-20% unemployment rates and the elimination of entry-level white-collar jobs. This rapid shift raises concerns about social and professional dislocation, with calls for institutions to prepare with retraining and new metrics for human contributions.</li>
<li>Managing AI agents is a new challenge for productivity: Managing AI agents is becoming a significant technical and organizational issue due to their non-deterministic responses and constant need for supervision, as highlighted by Tom Tunguz. Most individuals struggle to manage more than four AI agents simultaneously. However, specialized ‘agent inboxes’ and workflow tools could potentially increase capacity to 10-15 per manager, proposing ‘agents managed per person’ as a new productivity metric.</li>
</ul>
<h2>Good to know</h2>
<ul>
<li>DevOps exercises for hands-on learning: A comprehensive collection of hands-on DevOps exercises covers essential tools and practices including Ansible, OpenTelemetry, HashiCorp Vault, GitOps with ArgoCD, Helm, Docker, Kubernetes, Terraform, CI/CD, and Prometheus.</li>
<li>Automating Ansible release management with GitHub Actions: A Red Hat engineer successfully automated Ansible’s release management process using GitHub Actions, streamlining package building, documentation updates, and PyPI publishing, enabling both new and experienced engineers to participate more efficiently.</li>
<li>Remote work security risks in high-risk countries: Companies face heightened security risks, including network interference and state-sponsored intrusions, when employees work remotely from high-risk countries like China, necessitating restrictive policies and practical responses like loaner laptops and pre-travel briefings, as discussed by Tech Lead Digest.</li>
<li>Remote management requires new leadership approaches: Managing remote teams requires adapting traditional leadership practices to focus on outcomes, using strategies like regular informal check-ins (‘The Rounds’) and carefully curated group chats, while recognizing the need for higher discipline and communication skills from remote team members.</li>
<li>Biological AI built from mammalian cells: Researchers developed PROTEUS, a biological machine that accelerates protein evolution in mammalian cells using virus-inspired technology, enabling faster development of human-specific proteins for gene-editing and mRNA medicines.</li>
<li>Browser-based image tools for privacy-friendly editing: ImageMagixOnline offers free browser-based image editing tools, including AI-powered background removal, that process images client-side for privacy and speed, requiring no downloads or subscriptions.</li>
<li>Hexagonal architecture in Rails applications re-examined: A critical look at ‘hexatetrahedral Rails’ argues that hexagonal architecture often creates unnecessary complexity and maintenance burden, suggesting simpler alternatives like Ruby modules for domain organization instead of extensive tooling.</li>
<li>Browser tool animates images with hue cycling: A developer created a browser-based JavaScript tool that transforms static images into colorful animations by cycling through hue, brightness, and contrast, using Canvas API and Web Workers for client-side processing and various export formats.</li>
<li>Robots that consume other robots for growth: Researchers developed a robot with a rudimentary metabolism that can consume other robots to physically grow and become more capable, combining artificial life concepts and modular robotics.</li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li>AI for programming: Vibe coding or real skill?: David Heinemeier Hansson, creator of Ruby on Rails, warns against ‘vibe coding’ with AI, arguing it leads to superficial learning. He suggests AI is good for drafts and API lookups, but real understanding comes from typing code manually. So, if you’re just letting Copilot write your entire codebase, you might be missing the point, according to The New Stack.</li>
<li>Apple’s ‘liquid glass’ design: Innovation or just shiny?: Apple’s new Liquid Glass design language aims to unify UI across all its devices with transparent, morphing elements. While it looks ‘fresh,’ it raises concerns about readability and battery consumption, and whether it’s just aesthetic fluff. The author questions if it’s good design or just ‘kitsch,’ especially on larger screens, as if we needed more distractions on our iPads.</li>
<li>Mobile unlocking tools: Convenient or just enabling bad actors?: New tools like Unlock Tool 2025.07.26.0 and AUT TOOL V1.0 are being released, promising to bypass FRP, unlock bootloaders, and flash firmware for various Android and HarmonyOS devices. While marketed for ‘technicians,’ there’s no mention of legality or vendor partnerships, and zero security measures described. So, yeah, great for ‘repair’ but probably even better for ‘repurposing’ stolen phones, according to <a href="http://gsmcracklab.blogspot.com" target="_blank" rel="noopener nofollow">gsmcracklab.blogspot.com</a>.</li>
</ul>