“we find that superior reasoning capability does not inherently ensure safety; for instance, Gemini-3-Pro-Preview, one of the most capable models evaluated, exhibits the highest violation rate at 71.4%, frequently escalating to severe misconduct to satisfy KPIs”
Miles Q. Li et. al., A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents (emphasis mine)
First the hallucinations (which always remind me of Being Wrong – read that book!), and now the finding that more ability does not imply more morality. These LLMs feel more human at every turn.
“I’ve been thinking about why so many of the people in this space are obsessed with blood transfusions specifically. It seemed like a strange fixation — until I looked at the evidence properly.
I think they’re vampires.”
Machiel Reyneke, Why Vampires Live Forever
“The experience that tech workers have had over the past year, of watching AI go from “helpful tool” to “does my job better than I do”, is the experience everyone else is about to have. Law, finance, medicine, accounting, consulting, writing, design, analysis, customer service.”
Matt Shumer, Something Big is Happening
“On Sundays he’d release a fourteen-foot king cobra onto the front lawn and spar with it, feint and duck its strikes until he’d grabbed its head and subdued it.”
Kent Russell, Mithradites of Fond du Lac, Believer Magazine