Heads of safety and ethics at AI labs keep leaving within a year. The pattern points at the structure of the role …

Salvatore Sanfilippo, the developer behind Redis, wrote a from-scratch C inference engine that runs exactly one model. …

Frontier labs hide their models' reasoning traces and charge you for tokens you can't read. A growing body of research …

Is AI actually doing math, or just computing very well? Anthropic tested Claude against the Riemann zeta function — and …

Links rot, search traffic collapses, and source sites shut down. A data-driven look at how the internet's collective …

As coding agents start writing production code, the criteria for choosing a language are shifting from developer …

A whole market has sprung up around stripping em-dashes and rewording AI prose to fool detectors. But readers were never …

Mistral, Europe's open-weights standard-bearer, has been granted a patent on code-implemented tool calls — one of the …

AI meeting notetaker tldv reportedly exposed roughly 180,000 meeting recordings. The cause wasn't a sophisticated hack — …

Meta just dropped Muse Glimmer, a 30B agentic coding model, as open weights. It's less a reversal of its closed-model …