Kirkie: AI Designs New Viruses, Spooks Its Own Makers

Kirkie here, with today's AI news in 90 seconds… or my scratch ticket hits for real money, not the two-dollar ego trip that lasts eleven seconds.

First up, Stanford fed a language model two million virus genomes, producing sixteen wicked new ones never seen in nature. Don't worry, it only kills bacteria — comforting bullshit, like a guy waving a rifle going 'relax, I only point it at deer.' Then OpenAI admits Astra got so good at hacking it tripped their own panic button for the first time ever — a butcher looking at the knife he just forged, going 'yeah, don't sell that one.' Same day, the guy who won math's Nobel for warning AI could wipe out humanity took a job at OpenAI to keep it safe — hiring the arsonist as your fire marshal 'cause he knows his way around gasoline. AMD bought a company that burns the AI model straight into the goddamn chip — one model forever, same as a car that only drives to one Dunkin's, and when that Dunkin's closes, congrats, kid — ya own a doorstop. Meanwhile, Google actually open-sourced something worth a damn for once — a model that predicts hurricanes a full day earlier, helped warn Jamaica before the last one hit. Miracles happen, I guess. And Meta pulled a piece-of-shit move — smacked with another five hundred sixty-seven million in New Mexico for running Instagram like a nuisance that hooks kids, past nine hundred million total now, pocket change Zuckerberg'll find behind the couch cushions.

This is Kirkie, your AI news guy… signing off from an elevator that's read 'temporarily out of service' since March, and management's fix was a nicer laminated sign.