Don't Replace Links With Buttons
The post argues that using JavaScript buttons or popups as faux-links breaks browser features, accessibility, and user expectations, and recommends using real links and native controls so browsers and assistive tools work correctly. maurycyz.com
Hacky Way to Read errno
The author demonstrates a cross-language hack that captures stderr to read perror output and matches it against strerror strings to recover errno values without a C wrapper, noting the approach is nonstandard and brittle across libc implementations. sebsite.pw
Childlike Listening on AI Capability Arguments
The essay urges keeping a simple, first-order perspective—"childish listening"—to resist elaborate rationalizations for accelerating AI R&D and to ensure basic existential-risk concerns remain visible in decision‑making. lesswrong.com
AI Compute As a Tradable Asset
The Economist reports financial firms are racing to create instruments that let traders buy and sell processing power, framing AI compute as a new marketable commodity. economist.com
Plainspeak Dial for Gemma-4
A 300 KB control-vector (a "plainspeak" dial) for Gemma-4 reduces purple prose toward plainer human prose and, at the recommended setting, closes about 26% of the measured style gap without retraining. huggingface.co
Editorial Take
These five sources do not form a single coherent story beyond showing diverse technical and cultural friction points in modern software and AI work. Each piece is best read on its own merits. The link post is a plain usability and accessibility warning: prefer native link behavior to avoid breaking browser affordances. The errno write-up is an ingenious but fragile portability hack that trades safety for avoiding C wrappers. The LessWrong essay is a normative plea to preserve simple, first‑order intuitions about extreme risk even when complex rationales push for more capability work. The Economist brief signals a financialization of compute that could change incentives around who controls large-scale training resources. The Plainspeak dial is a lightweight, effective engineering control that shifts model register without retraining. Taken together they highlight recurring themes: prefer native primitives to hacks when possible; be wary of incentive shifts (whether markets or rationalizations) that can amplify risk; and look for small, well-scoped engineering controls that improve behavior without large systemic changes. This connects directly to the incentives and risk discussions in "AI Diffusion: Convergence, Risk, and Community Response."