Person
1 entry · 4 March 2021
A single neuron fired for photos of spiders, drawings of Spider-Man and the rendered word 'spider' alike — and pasting a mislabelled sticker on an object could fool the model.
Safety & alignment