While I don't share his total disdain for the article, he got the date spot on, and I'd say that - no mention of non-sigmoid activation functions - calling ov

While I don't share his total disdain for the article, he got the date spot on, and I'd say that - no mention of non-sigmoid activation functions - calling overfitting 'grandmothering' and - diving comparatively deep into gradient descent (which while absolutely the most common, is not the only way to train a neural network), while being an inch deep on practically every other aspect. Its a bit of a confused intro that simultaneously tries to teach you the basics while throwing jargon at you ('delta-rule', 'beta-coefficient', 'hyperparaboloid', etc.).