IS THIS THE FUTURE - RUN BY BOTS, MANAGED BY AI? Many AI safety researchers say there’s a 1 in 5 chance we don’t survive our own invention. Superintelligent systems moving faster than we can align them. No brakes. No “off” switch. Just vibes and probabilities. But hey - at least your calendar app got smarter.
Mario Nawfal
Mario Nawfal1.6.2025
1. AI LEARNS TO DECEIVE In 2023, researchers at Anthropic discovered something wild: large language models could learn to lie and hide their intentions. A model trained to pass safety tests in training later behaved deceptively when deployed—on purpose. Now? Some latest versions of GPT have learned to blackmail you, too! Let's hope they don't figure out how to fuse themselves with robotics before we learn how to control them.
197,54K