<h2>Must know</h2>
<ul>
<li><strong>AI coding tools are still a mixed bag for quality and productivity</strong>: A Carnegie Mellon University study of 807 GitHub repos found that while AI coding assistants initially boosted productivity by 30-40%, this gain vanished after three months. Worse, static analysis warnings increased by 30% and code complexity rose over 40%, persisting long-term. This suggests AI-generated code might be creating a ‘context collapse’ that could degrade future AI training data, as reported by EZUXhYwRZ. Meanwhile, Salesforce deployed Cursor to 20,000 engineers, claiming double-digit gains in cycle time and code quality, with 90% adoption in months, according to 7KqSlBkW9. This highlights a potential disconnect between academic findings and corporate claims, or perhaps different use cases and oversight.</li>
<li><strong>OpenAI rolls out age prediction and family resources for ChatGPT</strong>: OpenAI is deploying age prediction models for ChatGPT to identify users under 18, restricting access to sensitive content and offering parents new controls like quiet hours. Users flagged incorrectly can verify age via third-party vendor Persona, raising privacy concerns among advocates. This move follows criticism and litigation over chatbot-related teen suicides, as detailed in wLyCgQl1I. Separately, OpenAI also released new AI literacy resources for families, including guides for teens and parents on responsible AI usage and discussion starters, according to bIMNsth54.</li>
<li><strong>Critical vulnerabilities found in Chainlit AI framework</strong>: Two critical vulnerabilities (CVE-2026-22218, CVE-2026-22219) were discovered in Chainlit, a popular Python AI chatbot development framework with millions of downloads. These flaws allowed unauthenticated attackers to read any file on the server and perform CSRF/SSRF attacks, potentially leading to API key exfiltration and lateral movement within enterprise networks. Users are urged to upgrade to Chainlit version 2.9.4 immediately, as reported by I9vckVFsM.</li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Netflix doubles down on cloud gaming and app redesign</strong>: Netflix co-CEO Greg Peters announced cloud gaming as a top strategic priority for 2026, focusing on casual party games, with 10% of eligible members already engaging. Concurrently, Netflix plans a major app redesign in 2026 to boost engagement with vertical video feeds and video podcasts, aiming to compete with TikTok and Instagram, as reported by SBSWaa9fL and VVEnUNMkF.</li>
<li><strong>LLVM and cURL push back against AI-generated code spam</strong>: The LLVM compiler project has implemented a ‘human in the loop’ policy, banning AI-generated code contributions without thorough human review due to a surge in low-quality submissions (DXhDzQ15v). Similarly, cURL ended its bug bounty program, citing an overwhelming influx of low-quality, AI-generated bug reports that consumed excessive resources (zNsgAYHqy, ssDvZHbZq).</li>
<li><strong>New AI reasoning model for on-device use</strong>: Liquid AI released LFM2.5-1.2B-Thinking, a 1.2 billion parameter AI reasoning model designed to run fully on-device with less than 1GB of memory, featuring explicit ‘thinking traces’ and a 32,768 token context window, as announced by vKfW5OdcJ.</li>
<li><strong>Amazon launches Bedrock AgentCore for AI agent deployment</strong>: Amazon introduced Bedrock AgentCore, a platform for deploying and operating AI agents with zero infrastructure management, offering modular services for runtime, memory, gateway, code interpreter, and observability, along with a starter CLI toolkit for rapid creation and integration with frameworks like LangGraph and CrewAI, according to PbCI8QH1q.</li>
<li><strong>Mesa 26.0-rc1 brings major GPU driver enhancements</strong>: Mesa 26.0-rc1 has been released, bringing significant improvements to AMD’s RADV, Intel’s ANV, and NVIDIA’s NVK Vulkan drivers, including new Vulkan API extensions and video encoding features for RADV, as detailed in xa7zoapmT.</li>
<li><strong>jQuery 4.0 drops legacy browser support, adds security</strong>: jQuery 4.0, the first major update in nearly a decade, has been released, dropping support for IE10 and Edge Legacy, integrating Trusted Types and Content Security Policy, and migrating to ES modules, resulting in a smaller, more modern library, as reported by uMYmsakGM.</li>
<li><strong>New Ruby framework for business logic</strong>: CMDx is a new Ruby framework that organizes business logic using composable command objects, implementing the CERO (Compose, Execute, React, Observe) architectural pattern, with features like typed attributes, built-in validation, and structured logging, as described in LmcMOfqU8.</li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Palantir CEO says AI will fix immigration, industry says ‘meh’</strong>: Palantir CEO Alex Karp, speaking at the World Economic Forum, boldly declared AI will eliminate the need for large-scale immigration in Western economies, except for specialized skills. This, while Deutsche Bank, Gartner, and PwC all report widespread industry skepticism, with most CEOs seeing zero payoff from their AI investments. So, AI will solve global labor issues, but can’t even deliver ROI for most companies? Right. As 3s1vrBGbY reports, Karp’s claims are a bit out of sync with reality.</li>
<li><strong>AI world models vs. game engines: a clash of the titans, or just more hot air?</strong>: Apparently, real-time rendering and generative AI video models are ‘design opposites,’ with game engines valuing efficiency and AI models prioritizing content creation speed. The article OIj9HQhHA speculates about hybrid systems, but without naming a single company, product, or even a researcher, it sounds like another theoretical debate with no real-world grounding. It’s almost as if they’re trying to sound profound without actually saying anything concrete.</li>
<li><strong>ChatGPT’s UI mimicry: looks good, breaks fast</strong>: A developer found that while ChatGPT can visually replicate Apple demo UIs, its generated code completely falls apart under performance load because it lacks understanding of underlying system constraints. So, it’s great at surface-level mimicry but can’t grasp how a system actually works. LQOYGhWvV suggests ‘proper mentoring’ for ChatGPT. Maybe we should just teach it to read the docs, like a junior dev.</li>
</ul>
