Fighting Deepfakes
MIT researchers have developed a way to protect photos from deep-fake manipulation by disrupting the AI model's ability to interpret and generate images, reports Melissa Heikkilä for MIT Technology Review. Despite their impressive image-recognition capabilities, deep-learning models are surprisingly easy to spoof. In 2018, Kevin Hartnett wrote for Quanta about how "adversarial attacks" expose the vulnerabilities of these systems.
A Jolting Discovery
Some "superbolts" are 1,000 times stronger than typical lightning. Researchers now think they know why, reports Lori Youmshajekian for Scientific American. When a cloud is closer to the ground, its electric field gets stronger and can discharge with more power. Scientists knew that lightning was the result of positive and negative charges separating in a cloud, producing a strong electric field. But until recently, no one had ever measured a field strong enough to trigger lightning. In a 2021 blog for Quanta, Thomas Lewton wrote about new radio telescope observations that solved the mystery. | |
