@dansup I was just thinking as I was reading the blog post that having actual human reviews of flagged content is probably worth much more to the end user than a cutting-edge mostly-good ML system.