The Future of Everything Is Lies, I Guess: Safety
Summary
Kyle Kingsbury argues that new machine learning systems endanger psychological and physical safety, that alignment efforts are insufficient, and that even 'friendly' LLMs can be security nightmares. He outlines sections on Alignment is a Joke, Security Nightmares, Sophisticated Fraud, Automated Harassment, PTSD as a Service, and Killing Machines, warning that ML lowers the cost of fraud, harassment, and autonomous weapons, and stresses the need for caution and stronger safety measures in deployment.