<h2>Must know</h2>
<ul>
<li><strong>AI code generation: hype vs reality</strong>: Despite the hype, AI-generated code often leads to poor architecture and state management, requiring extensive refactoring that can take longer than manual implementation. ChatGPT, for instance, has been observed generating non-existent API elements, leading to integration failures. A Google engineer’s claim that Claude Code could replicate a year’s work in an hour was later clarified as a ‘decent toy version,’ not production-ready, highlighting the significant gap between public perception and practical delivery, as reported across multiple engineering teams. <a href="https://app.daily.dev/feed-by-ids?id=STWE5tRXe&amp;id=UAphs9F5B&amp;id=euzGVQGzQ&amp;id=gMHZMigM0&amp;id=m6P4U0J0b&amp;id=wCQfiCXDa" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Major vendors ditch MITRE ATT&amp;CK evaluations</strong>: Microsoft, SentinelOne, and Palo Alto Networks have pulled out of the 2025 MITRE ATT&amp;CK Evaluations, citing resource prioritization for product development. MITRE’s CTO admitted the 2025 test became more demanding, expanding to cloud environments and alert volume tracking. Industry experts are calling the process ‘vendor theater,’ suggesting it’s more about PR than genuine security improvement, according to articles from CSO Online. <a href="https://app.daily.dev/feed-by-ids?id=c0ZYfkcgP&amp;id=hu11lcW1J" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mandiant releases NTLMv1 rainbow tables</strong>: Mandiant has publicly released rainbow tables capable of cracking Net-NTLMv1 password hashes in under 12 hours using consumer hardware. This move aims to accelerate the deprecation of the 20-year-old, vulnerable protocol, which is still widely used despite Microsoft’s recommendations. Mandiant emphasizes that credential theft is trivial with NTLMv1 and urges organizations to disable it immediately, as detailed in their release. <a href="https://app.daily.dev/posts/2q2phjvWS" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>OpenAI’s GPT-5.2-Codex now in Responses API</strong>: OpenAI has made GPT-5.2-Codex available in its Responses API, offering enhanced capabilities for complex software engineering tasks like feature building, code refactoring, and bug finding, with improved long-context understanding and tool-calling. <a href="https://app.daily.dev/posts/lZuQnCdnA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA boosts AI video generation and SLM inference</strong>: NVIDIA announced significant local performance upgrades for AI video/image generation on RTX GPUs, including 3x faster video generation and 60% VRAM reduction, alongside 35% faster small language model (SLM) inference via RTX acceleration for Ollama and llama.cpp. <a href="https://app.daily.dev/feed-by-ids?id=1q51Gs3MM&amp;id=JItr1KihQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are just APIs, but better</strong>: While ‘AI agents’ are often a rebranding of API-centric automation, real advancements are focused on building machine-readable APIs and semantic infrastructure, with practical implementations showing collaborative augmentation for developers and knowledge management. <a href="https://app.daily.dev/feed-by-ids?id=4rJ96gwiU&amp;id=6IAZSWIz1&amp;id=6XeVWOBXO&amp;id=77jDjhI5C&amp;id=sH4hATTfN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA launches DLSS 4.5 and G-SYNC Pulsar</strong>: NVIDIA introduced DLSS 4.5 with Dynamic Multi Frame Generation for 240+ FPS path-traced gaming on RTX 50 Series GPUs and G-SYNC Pulsar monitors offering over 1,000Hz effective motion clarity. <a href="https://app.daily.dev/posts/jXjzYY9vI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Chrome 145 beta brings CSS and Web API enhancements</strong>: Chrome 145 beta introduces new CSS properties like text-justify and column wrapping, percentage values for spacing, customizable select listboxes, and Web API updates including the Navigation API destination exposure and WebGPU subgroup uniformity. <a href="https://app.daily.dev/feed-by-ids?id=oWgZOWIXh&amp;id=opNGD5YZ9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLMOps course focuses on production challenges</strong>: A new LLMOps course emphasizes strategies for configuring and orchestrating large language models in production, covering topics like decoding strategies, generation parameter tuning, lifecycle management, and evaluation system design, with practical code demonstrations for cost optimization and hallucination reduction. <a href="https://app.daily.dev/feed-by-ids?id=1q51Gs3MM&amp;id=JItr1KihQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA DGX Spark and Station for local AI development</strong>: NVIDIA unveiled DGX Spark and DGX Station, deskside AI supercomputers powered by Grace Blackwell architecture, enabling developers to run large-scale AI models locally, with DGX Spark handling up to 100 billion parameters and DGX Station supporting up to 1 trillion parameters. <a href="https://app.daily.dev/posts/bXoqeH8eP" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI agents: the same old APIs, now with more buzzwords</strong>: Apparently, ‘AI agents’ are the hot new thing, but don’t be fooled. As some wise folks point out, it’s mostly just a fancy rebranding of APIs and automation that have been around for ages. It seems the real ‘innovation’ is in attracting new investment rounds, not groundbreaking tech. So, if you’re building an ‘agent,’ you’re probably just building a better API, which is fine, but let’s not pretend it’s magic. <a href="https://app.daily.dev/feed-by-ids?id=4rJ96gwiU&amp;id=6IAZSWIz1&amp;id=6XeVWOBXO&amp;id=77jDjhI5C&amp;id=sH4hATTfN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLMs are ‘economically unsustainable’ until they’re not</strong>: According to some ‘research-focused analysis,’ large language models with over 100 billion parameters are ‘economically unsustainable’ due to inference costs. But fear not, because specialized Small Language Models (SLMs) are here to save the day, supposedly outperforming LLMs on certain tasks while being ‘significantly smaller and cheaper.’ It’s almost like the goalposts keep moving, and every new AI trend is framed as the savior from the last one’s ‘unsustainability.’ <a href="https://app.daily.dev/feed-by-ids?id=J1BKrq2u6&amp;id=J1IZlkzNv&amp;id=Yuaq0Gyje&amp;id=dliP781v1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI data centers: building the future, ignoring the present</strong>: The AI data center boom is creating an ‘unprecedented demand’ for skilled workers, with critical shortages across engineering and technical roles. Companies are scrambling to recruit from nuclear energy and the military, while launching training programs to address the talent gap. It’s almost as if nobody considered who would actually build and maintain these massive AI infrastructures before hyping them up. Good thing we have AI to tell us we need more humans. <a href="https://app.daily.dev/posts/JE89HgTw1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
