Neural networks are not already clean

A LessWrong article argues that neural networks, despite common beliefs, are not inherently clean or free from biases and issues. It challenges the assumption that neural networks naturally avoid problems without deliberate intervention.