Marginalium

AI and Intelligence Analysis. Generally interesting, though they point out the data issue. And also suggest it’ll help avoid cognitive bias, though obviously we’d just be replacing one set of biases with another (whatever the model was trained on).